{ "cells": [ { "cell_type": "code", "execution_count": 1, "id": "bf70f34f", "metadata": {}, "outputs": [], "source": [ "import os\n", "import re\n", "import json\n", "import tqdm\n", "import hashlib\n", "import uuid\n", "import random\n", "import funcy\n", "import copy\n", "import shutil\n", "import portion\n", "from collections import defaultdict\n", "import numpy as np\n", "import pandas as pd\n", "\n", "from suno_utils.tasks.music import remove_music\n", "from suno_utils.audio import Audio, Token, Tokens\n", "from suno_utils.audio.conversion import get_audio_properties, play_audio, get_duration_s\n", "from suno_utils.utils.numbers import safe_round\n", "from suno_utils.utils.text import normalize_whitespace\n", "from suno_utils.web.harvest import get_file_ext, get_filename\n", "from suno_utils.utils.slicer import get_vad_gaps, get_intervals_from_gaps\n", "\n", "SAMPLE_RATE = 16_000\n", "\n", "BASE_DIR = \"/mnt/data-ssd-1/data/private/customer/level_ai/pilot/\"\n", "RAW_DATA_DIR = os.path.join(BASE_DIR, \"raw_data\")\n", "RAW_AUDIO_DIR = os.path.join(RAW_DATA_DIR, \"audio\")\n", "\n", "AUDIO_DIR = os.path.join(BASE_DIR, \"audio\")\n", "TO_REV_DIR = os.path.join(BASE_DIR, \"to_rev\")\n", "TO_REV_AUDIO_SEGMENTS_DIR = os.path.join(TO_REV_DIR, \"audio_segments\")\n", "TO_REV_AUDIO_DIR = os.path.join(TO_REV_DIR, \"audio\")\n", "FROM_REV_DIR = os.path.join(BASE_DIR, \"from_rev\")\n", "ARTIFACT_DIR = os.path.join(BASE_DIR, \"artifacts\")\n", "\n", "PIPELINE_DATE_STR = \"2022_10_20\"\n", "PIPELINE_SEGMENT_DIR = os.path.join(BASE_DIR, \"pipeline\", PIPELINE_DATE_STR, \"segments\")\n", "PIPELINE_ARTIFACT_DIR = os.path.join(BASE_DIR, \"pipeline\", PIPELINE_DATE_STR, \"artifacts\")" ] }, { "cell_type": "code", "execution_count": 2, "id": "c769f21a", "metadata": {}, "outputs": [], "source": [ "os.makedirs(AUDIO_DIR, exist_ok=True)\n", "os.makedirs(TO_REV_DIR, exist_ok=True)\n", "os.makedirs(TO_REV_AUDIO_SEGMENTS_DIR, exist_ok=True)\n", "os.makedirs(TO_REV_AUDIO_DIR, exist_ok=True)\n", "os.makedirs(FROM_REV_DIR, exist_ok=True)\n", "os.makedirs(ARTIFACT_DIR, exist_ok=True)\n", "os.makedirs(PIPELINE_SEGMENT_DIR, exist_ok=True)\n", "os.makedirs(PIPELINE_ARTIFACT_DIR, exist_ok=True)" ] }, { "cell_type": "code", "execution_count": 3, "id": "c3cd79ed", "metadata": {}, "outputs": [], "source": [ "GLOBAL_HOTWORDS = []" ] }, { "cell_type": "markdown", "id": "3bf11bab", "metadata": {}, "source": [ "## Process raw data" ] }, { "cell_type": "code", "execution_count": 17, "id": "6a80e561", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "12 files processed.\n", "1.0 hours of audio.\n" ] } ], "source": [ "# save in dir with known sample rate as wavs\n", "filepaths_info = []\n", "for fn in os.listdir(RAW_AUDIO_DIR):\n", " try:\n", " audio = Audio.from_file(os.path.join(RAW_AUDIO_DIR, fn), sample_rate=SAMPLE_RATE, byte_width=2)\n", " except:\n", " print(f\"could not convert: {fn}\")\n", " uid = str(uuid.uuid4())\n", " from_fp = os.path.join(RAW_AUDIO_DIR, fn)\n", " to_fp = os.path.join(AUDIO_DIR, fn)\n", " to_filepath = os.path.join(AUDIO_DIR, uid + \".wav\")\n", " audio.to_wav(to_filepath)\n", " filepaths_info.append({\n", " \"original_filename\": fn,\n", " \"uid\": uid,\n", " \"filepath\": to_filepath,\n", " \"duration_s\": audio.duration_s,\n", " })\n", "print(f\"{len(filepaths_info)} files processed.\")\n", "print(round(np.sum([m[\"duration_s\"] for m in filepaths_info]) / 60 / 60, 1), \"hours of audio.\")" ] }, { "cell_type": "code", "execution_count": 18, "id": "deb0bebc", "metadata": {}, "outputs": [], "source": [ "with open(os.path.join(ARTIFACT_DIR, \"raw_data_manifest.json\"), \"w\") as f:\n", " json.dump(filepaths_info, f)" ] }, { "cell_type": "markdown", "id": "ab10131a", "metadata": {}, "source": [ "## Prepare rev data" ] }, { "cell_type": "code", "execution_count": 136, "id": "f7f441a5", "metadata": {}, "outputs": [], "source": [ "# sub-segment if necessary\n", "def _segment_audio(\n", " audio, \n", " min_segment_dur_s=3, \n", " max_segment_dur_s=15, \n", " min_silence_gap_s=0.1,\n", " vad_level=3,\n", " clean_audio=False, \n", " remove_silence=False, \n", " keep_all_lengths=True,\n", "):\n", " # TODO: remove background noise?\n", " if clean_audio:\n", " cleaned_audio = remove_music(audio, use_gpu=True)\n", " else:\n", " cleaned_audio = audio\n", " # TODO: automatically tune vad_level and min gap size?\n", " vad_gaps = get_vad_gaps(cleaned_audio, min_gap_s=min_silence_gap_s, vad_level=vad_level, silent=True)\n", " silence_frac = np.sum([g.upper - g.lower for g in vad_gaps]) / audio.duration_s\n", "# print(\"{}% silence\".format(round(silence_frac * 100, 1)))\n", "\n", " # anything longer than a second should get excluded with pad\n", " silence_intervals = portion.empty()\n", " if remove_silence:\n", " max_silence_s = 1\n", " max_silence_pad = max_silence_s / 2\n", " for vad_gap in vad_gaps:\n", " excluded_lower_s = safe_round(vad_gap.lower + max_silence_pad)\n", " excluded_upper_s = safe_round(vad_gap.upper - max_silence_pad)\n", " if excluded_upper_s - excluded_lower_s > 0:\n", " silence_intervals |= portion.openclosed(excluded_lower_s, excluded_upper_s)\n", " \n", " reliable_speech_segments = get_intervals_from_gaps(\n", " vad_gaps,\n", " excluded_intervals=silence_intervals,\n", " min_dur_s=min_segment_dur_s,\n", " max_dur_s=max_segment_dur_s,\n", " total_duration_s=audio.duration_s,\n", " )\n", "\n", " tot_segment_duration_s = np.sum([end_s - start_s for start_s, end_s in reliable_speech_segments])\n", " cov_frac = tot_segment_duration_s / audio.duration_s\n", "\n", " # add non-silent speech fragments\n", " speech_segments = reliable_speech_segments[:]\n", " if keep_all_lengths:\n", " speech_segment_interval = portion.empty()\n", " for start_s, end_s in reliable_speech_segments:\n", " speech_segment_interval |= portion.closedopen(start_s, end_s)\n", " non_covered_speech = portion.openclosed(0, audio.duration_s) - silence_intervals - speech_segment_interval\n", " if not non_covered_speech.empty:\n", " for s in non_covered_speech:\n", " offs_s = safe_round(s.lower)\n", " while True:\n", " end_s = safe_round(np.min([s.upper, offs_s + max_segment_dur_s]))\n", " if end_s - offs_s <= 0:\n", " break\n", " speech_segments.append((offs_s, end_s))\n", " offs_s = end_s\n", " speech_segments = sorted(speech_segments)\n", "\n", " tot_segment_duration_s = np.sum([end_s - start_s for start_s, end_s in speech_segments])\n", " if np.abs(audio.duration_s - tot_segment_duration_s) > 0.5:\n", " raise ValueError(\"only {}% segment coverage\".format(round(cov_frac * 100, 1)))\n", " \n", " segment_metas = [\n", " {\n", " \"start_s\": start_s,\n", " \"end_s\": end_s,\n", " } for start_s, end_s in speech_segments\n", " ]\n", " return segment_metas" ] }, { "cell_type": "code", "execution_count": 57, "id": "ff3fb41d", "metadata": {}, "outputs": [], "source": [ "shutil.rmtree(TO_REV_AUDIO_SEGMENTS_DIR, ignore_errors=True)\n", "shutil.rmtree(TO_REV_AUDIO_DIR, ignore_errors=True)\n", "os.makedirs(TO_REV_AUDIO_SEGMENTS_DIR, exist_ok=True)\n", "os.makedirs(TO_REV_AUDIO_DIR, exist_ok=True)" ] }, { "cell_type": "code", "execution_count": 58, "id": "d123ad24", "metadata": {}, "outputs": [], "source": [ "with open(os.path.join(ARTIFACT_DIR, \"raw_data_manifest.json\")) as f:\n", " filepaths_info = json.load(f)" ] }, { "cell_type": "code", "execution_count": 59, "id": "fb8a0837", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "{'original_filename': 'REc88644a5aa901bad8e1d889c9fe3849a.wav',\n", " 'uid': '6ee449d1-74a5-435b-a274-071196f4f3b7',\n", " 'filepath': '/mnt/data-ssd-1/data/private/customer/level_ai/pilot/audio/6ee449d1-74a5-435b-a274-071196f4f3b7.wav',\n", " 'duration_s': 198.9905}" ] }, "execution_count": 59, "metadata": {}, "output_type": "execute_result" } ], "source": [ "m = filepaths_info[0]\n", "m" ] }, { "cell_type": "code", "execution_count": 45, "id": "11ab6f9d", "metadata": {}, "outputs": [], "source": [ "# TODO: do we want to normalize the volume?\n", "# from pydub import AudioSegment, effects \n", "\n", "# audio_seg = AudioSegment(\n", "# data=audio.array, \n", "# frame_rate=audio.sample_rate, \n", "# sample_width=audio.byte_width, \n", "# channels=audio.n_channels,\n", "# )\n", "# norm_seg = effects.normalize(audio_seg) \n", "# norm_audio = Audio(norm_seg.raw_data, 16_000, 2, 1)" ] }, { "cell_type": "code", "execution_count": 60, "id": "4da818da", "metadata": { "scrolled": true }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "43.9 mins of audio (9.7% overhead)\n", "22.5 mins of audio (9.8% overhead)\n" ] } ], "source": [ "MIN_SEGMENT_DURATION_S = 3\n", "MAX_SEGMENT_DURATION_S = 15 # None for no sementing\n", "TARGET_REV_FILE_DURATION_S = 40 * 60\n", "MAX_REV_FILE_DURATION_S = 90 * 60\n", "INDICATOR_AUDIO_DIR = \"/mnt/data-ssd-1/data/custom/segment_indicators\"\n", "SILENCE_DURATION_S = 0.25\n", "\n", "audio_rev_metas = []\n", "tmp_segments_audio = []\n", "tmp_segments_meta = []\n", "tmp_n = 0\n", "for n_step, m in enumerate(filepaths_info):\n", " # load and subsegment audio if necessary\n", " audio = Audio.from_file(m[\"filepath\"], sample_rate=SAMPLE_RATE, byte_width=2)\n", " if MAX_SEGMENT_DURATION_S is not None and audio.duration_s > MAX_SEGMENT_DURATION_S:\n", " segment_metas = _segment_audio(\n", " audio, \n", " min_segment_dur_s=MIN_SEGMENT_DURATION_S, \n", " max_segment_dur_s=MAX_SEGMENT_DURATION_S, \n", " min_silence_gap_s=0.1,\n", " vad_level=3,\n", " clean_audio=False, \n", " remove_silence=False, \n", " keep_all_lengths=True,\n", " )\n", " else:\n", " segment_metas = [{\"start_s\": None, \"end_s\": None}]\n", " # assemble rev audio\n", " for mm in segment_metas:\n", " # add indicator\n", " indicator_str = str(tmp_n).zfill(3)\n", " tmp_indicator_audios = []\n", " for c in indicator_str:\n", " tmp_indicator_audios.append(\n", " Audio.from_file(\n", " os.path.join(INDICATOR_AUDIO_DIR, f\"{c}_fast.wav\"), \n", " sample_rate=SAMPLE_RATE, \n", " byte_width=2,\n", " )\n", " )\n", " indicator_audio = Audio.concatenate(tmp_indicator_audios)\n", " tmp_segments_audio.append(indicator_audio)\n", " tmp_segments_meta.append({\n", " \"segment_number\": tmp_n,\n", " \"type\": \"indicator\",\n", " \"duration_s\": indicator_audio.duration_s,\n", " \"indicator_str\": indicator_str,\n", " })\n", " # add silence\n", " if SILENCE_DURATION_S > 0:\n", " silence_audio = Audio.from_array(\n", " np.zeros(int(SAMPLE_RATE * SILENCE_DURATION_S), dtype=np.int16), SAMPLE_RATE\n", " )\n", " tmp_segments_audio.append(silence_audio)\n", " tmp_segments_meta.append({\n", " \"segment_number\": tmp_n,\n", " \"type\": \"silence\",\n", " \"duration_s\": SILENCE_DURATION_S,\n", " })\n", " # add audio\n", " segment_audio = audio.get_slice(from_s=mm[\"start_s\"], to_s=mm[\"end_s\"])\n", " segment_uid = str(uuid.uuid4())\n", " segment_filepath = os.path.join(TO_REV_AUDIO_SEGMENTS_DIR, f\"{segment_uid}.mp3\")\n", " segment_audio.to_mp3(segment_filepath)\n", " tmp_segments_audio.append(segment_audio)\n", " tmp_segments_meta.append({\n", " \"segment_number\": tmp_n,\n", " \"type\": \"speech\",\n", " \"duration_s\": segment_audio.duration_s,\n", " \"id\": segment_uid,\n", " \"filepath\": segment_filepath,\n", " \"original_audio_id\": m[\"uid\"],\n", " \"original_audio_filepath\": m[\"filepath\"],\n", " \"original_audio_start_s\": mm[\"start_s\"],\n", " \"original_audio_end_s\": mm[\"end_s\"],\n", " })\n", " # add silence\n", " if SILENCE_DURATION_S > 0:\n", " silence_audio = Audio.from_array(\n", " np.zeros(int(SAMPLE_RATE * SILENCE_DURATION_S), dtype=np.int16), SAMPLE_RATE\n", " )\n", " tmp_segments_audio.append(silence_audio)\n", " tmp_segments_meta.append({\n", " \"segment_number\": tmp_n,\n", " \"type\": \"silence\",\n", " \"duration_s\": SILENCE_DURATION_S,\n", " })\n", " tmp_n += 1 \n", " # prep for next step and decide if we make new file\n", " running_duration_s = np.sum([e[\"duration_s\"] for e in tmp_segments_meta])\n", " if running_duration_s > MAX_REV_FILE_DURATION_S:\n", " raise ValueError(\"too long, rev might make diarization errors\")\n", " if (\n", " (running_duration_s >= TARGET_REV_FILE_DURATION_S) or \n", " (n_step == len(filepaths_info) - 1 and running_duration_s > 0)\n", " ):\n", " audio_rev = Audio.concatenate(tmp_segments_audio)\n", " running_speech_duration_s = np.sum([e[\"duration_s\"] for e in tmp_segments_meta if e[\"type\"] == \"speech\"])\n", " print(\"{} mins of audio ({}% overhead)\".format(\n", " round(running_duration_s / 60, 1),\n", " round((1 - running_speech_duration_s / running_duration_s) * 100, 1)\n", " )) \n", " uid = str(uuid.uuid4())\n", " rev_audio_filepath = os.path.join(TO_REV_AUDIO_DIR, f\"{uid}.mp3\")\n", " audio_rev.to_mp3(rev_audio_filepath)\n", " audio_rev_metas.append({\n", " \"uid\": uid,\n", " \"filepath\": rev_audio_filepath,\n", " \"segments_meta\": tmp_segments_meta,\n", " \"duration_s\": audio_rev.duration_s,\n", " })\n", " tmp_segments_audio = []\n", " tmp_segments_meta = []\n", " tmp_n = 0" ] }, { "cell_type": "code", "execution_count": 17, "id": "29f12158", "metadata": {}, "outputs": [], "source": [ "# # verify duration\n", "# l = []\n", "# for m in audio_rev_metas:\n", "# for mm in m[\"segments_meta\"]:\n", "# if mm[\"type\"] != \"speech\":\n", "# continue\n", "# l.append({\n", "# \"start_s\": mm[\"original_audio_start_s\"],\n", "# \"end_s\": mm[\"original_audio_end_s\"],\n", "# })\n", "# print(round(\n", "# np.sum([m[\"end_s\"] - m[\"start_s\"] for m in l]) / 60 / 60, 1\n", "# ), \"hours\")" ] }, { "cell_type": "code", "execution_count": 64, "id": "61ddadf4", "metadata": {}, "outputs": [], "source": [ "# !ls /mnt/data-ssd-1/data/private/customer/level_ai/pilot/to_rev/audio_segments" ] }, { "cell_type": "code", "execution_count": 65, "id": "c25676a9", "metadata": {}, "outputs": [], "source": [ "# !ls /mnt/data-ssd-1/data/private/customer/level_ai/pilot/to_rev/audio" ] }, { "cell_type": "code", "execution_count": 70, "id": "4d16d48e", "metadata": {}, "outputs": [], "source": [ "with open(os.path.join(TO_REV_DIR, f\"audio_rev_metas.json\"), \"w\") as f:\n", " json.dump(audio_rev_metas, f)" ] }, { "cell_type": "code", "execution_count": 123, "id": "b494b66c", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "[{'segment_number': 34,\n", " 'type': 'speech',\n", " 'duration_s': 14.41,\n", " 'id': 'd4b8614e-d8c9-4ca0-b25e-53c948a30382',\n", " 'filepath': '/mnt/data-ssd-1/data/private/customer/level_ai/pilot/to_rev/audio_segments/d4b8614e-d8c9-4ca0-b25e-53c948a30382.mp3',\n", " 'original_audio_id': '5ce83f7b-6b4b-43a5-af4e-eacfae3ebaad',\n", " 'original_audio_filepath': '/mnt/data-ssd-1/data/private/customer/level_ai/pilot/audio/5ce83f7b-6b4b-43a5-af4e-eacfae3ebaad.wav',\n", " 'original_audio_start_s': 285.4,\n", " 'original_audio_end_s': 299.81},\n", " {'segment_number': 35,\n", " 'type': 'speech',\n", " 'duration_s': 15.0,\n", " 'id': '085f33b5-8156-4973-b007-f9825f060b11',\n", " 'filepath': '/mnt/data-ssd-1/data/private/customer/level_ai/pilot/to_rev/audio_segments/085f33b5-8156-4973-b007-f9825f060b11.mp3',\n", " 'original_audio_id': '5ce83f7b-6b4b-43a5-af4e-eacfae3ebaad',\n", " 'original_audio_filepath': '/mnt/data-ssd-1/data/private/customer/level_ai/pilot/audio/5ce83f7b-6b4b-43a5-af4e-eacfae3ebaad.wav',\n", " 'original_audio_start_s': 299.81,\n", " 'original_audio_end_s': 314.81}]" ] }, "execution_count": 123, "metadata": {}, "output_type": "execute_result" } ], "source": [ "[m for m in audio_rev_metas[0][\"segments_meta\"] if m[\"type\"] == \"speech\"][34:36]" ] }, { "cell_type": "code", "execution_count": null, "id": "b0711135", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "a5dea8f6", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "5fc18307", "metadata": {}, "source": [ "## (optional) investigate" ] }, { "cell_type": "code", "execution_count": 72, "id": "dc409c9d", "metadata": {}, "outputs": [], "source": [ "with open(os.path.join(TO_REV_DIR, f\"audio_rev_metas.json\")) as f:\n", " audio_rev_metas = json.load(f)" ] }, { "cell_type": "code", "execution_count": 74, "id": "3604b3cb", "metadata": {}, "outputs": [ { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "Audio.from_file(audio_rev_metas[0][\"filepath\"]).play()" ] }, { "cell_type": "code", "execution_count": 75, "id": "40f5c4a2", "metadata": {}, "outputs": [ { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "Audio.from_file(audio_rev_metas[-1][\"filepath\"]).play()" ] }, { "cell_type": "code", "execution_count": null, "id": "92a8e1ba", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "491772f5", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "426ba2c3", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "dfc212cd", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "88a9e9cc", "metadata": {}, "source": [ "## Send stuff to rev" ] }, { "cell_type": "code", "execution_count": 76, "id": "05fbe672", "metadata": {}, "outputs": [], "source": [ "from suno_utils.web.rev import get_auth_string, upload_file, make_order, make_cc_order\n", "\n", "with open(\"/home/georg/.secrets/secrets.json\") as f:\n", " secrets = json.load(f)\n", " \n", "CLIENT_API_KEY = secrets[\"REV_CLIENT_API_KEY\"]\n", "USER_API_KEY = secrets[\"REV_USER_API_KEY\"]\n", "AUTH_STR = get_auth_string(CLIENT_API_KEY, USER_API_KEY)" ] }, { "cell_type": "code", "execution_count": 77, "id": "2db71de1", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "1.1 hours total\n" ] } ], "source": [ "with open(os.path.join(TO_REV_DIR, f\"audio_rev_metas.json\")) as f:\n", " audio_rev_metas = json.load(f)\n", "\n", "ORDER_REF_STR = \"lvl_poc\"\n", "\n", "tot_duration_s = np.sum([m[\"duration_s\"] for m in audio_rev_metas])\n", "print(round(tot_duration_s / 60 / 60, 1), \"hours total\")" ] }, { "cell_type": "code", "execution_count": 78, "id": "8c66f7f9", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:06<00:00, 3.12s/it]\n" ] } ], "source": [ "order_items = []\n", "for m in tqdm.tqdm(audio_rev_metas):\n", " media_loc = upload_file(AUTH_STR, m[\"filepath\"], file_ref_str=m[\"uid\"])\n", " order_items.append(\n", " {\n", " \"media_loc\": media_loc,\n", " \"hotwords\": GLOBAL_HOTWORDS,\n", " \"duration_s\": int(np.ceil(m[\"duration_s\"])),\n", " \"uuid\": m[\"uid\"],\n", " }\n", " )" ] }, { "cell_type": "code", "execution_count": 79, "id": "a0ee5f4f", "metadata": {}, "outputs": [], "source": [ "with open(os.path.join(TO_REV_DIR, \"01_order_items.json\"), \"w\") as f:\n", " json.dump(order_items, f)" ] }, { "cell_type": "code", "execution_count": 80, "id": "8318e46f", "metadata": {}, "outputs": [], "source": [ "order_number = make_order(AUTH_STR, order_items, order_ref_str=ORDER_REF_STR)\n", "# order_number = make_cc_order(AUTH_STR, order_items, order_ref_str=ORDER_REF_STR)" ] }, { "cell_type": "code", "execution_count": 81, "id": "27549d51", "metadata": {}, "outputs": [], "source": [ "with open(os.path.join(TO_REV_DIR, \"02_order_numer.txt\"), \"w\") as f:\n", " f.write(order_number)" ] }, { "cell_type": "code", "execution_count": null, "id": "4b13bb34", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "576b2bef", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "b33dc8e0", "metadata": {}, "source": [ "## Get rev stuff" ] }, { "cell_type": "code", "execution_count": 82, "id": "e8b59ece", "metadata": {}, "outputs": [], "source": [ "from suno_utils.web.rev import get_auth_string, get_finished_order, get_transcript\n", "\n", "with open(\"/home/georg/.secrets/secrets.json\") as f:\n", " secrets = json.load(f)\n", " \n", "CLIENT_API_KEY = secrets[\"REV_CLIENT_API_KEY\"]\n", "USER_API_KEY = secrets[\"REV_USER_API_KEY\"]\n", "AUTH_STR = get_auth_string(CLIENT_API_KEY, USER_API_KEY)" ] }, { "cell_type": "code", "execution_count": 83, "id": "9e4e40af", "metadata": {}, "outputs": [], "source": [ "with open(os.path.join(TO_REV_DIR, \"01_order_items.json\")) as f:\n", " order_items = json.load(f)\n", "with open(os.path.join(TO_REV_DIR, \"02_order_numer.txt\")) as f:\n", " order_number = f.read()" ] }, { "cell_type": "code", "execution_count": 84, "id": "c4d6d755", "metadata": {}, "outputs": [], "source": [ "# # grab while still in progress\n", "# from suno_utils.web.rev import _get_order_details, _parse_order_response\n", "# order_details = _get_order_details(AUTH_STR, order_number)\n", "# _, transcript_metas = _parse_order_response(order_details)\n", "# print(f\"{len(transcript_metas)}/{len(order_items)} transcripts pulled\")" ] }, { "cell_type": "code", "execution_count": 85, "id": "9275ae0d", "metadata": {}, "outputs": [], "source": [ "# grab all of it\n", "transcript_metas = get_finished_order(AUTH_STR, order_number, blocking=False)" ] }, { "cell_type": "code", "execution_count": 86, "id": "df28d887", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:02<00:00, 1.31s/it]\n" ] } ], "source": [ "# get json\n", "for transcript_meta in tqdm.tqdm(transcript_metas):\n", " transcript_filepath = os.path.join(FROM_REV_DIR, transcript_meta[\"uuid\"] + \".json\")\n", " if os.path.exists(transcript_filepath):\n", " continue\n", " transcript_json = get_transcript(AUTH_STR, transcript_meta)\n", " with open(transcript_filepath, \"w\") as f:\n", " json.dump(transcript_json, f)" ] }, { "cell_type": "code", "execution_count": null, "id": "8fc2a35d", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "4d3935ec", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "9be84dd0", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "9cffb6b3", "metadata": {}, "source": [ "## Make manifest and run pipeline" ] }, { "cell_type": "code", "execution_count": 72, "id": "7d75b028", "metadata": {}, "outputs": [], "source": [ "with open(os.path.join(TO_REV_DIR, f\"audio_rev_metas.json\")) as f:\n", " audio_rev_metas = json.load(f)" ] }, { "cell_type": "code", "execution_count": 73, "id": "db8a1a08", "metadata": {}, "outputs": [], "source": [ "# make sure to delete segments when rerunning\n", "shutil.rmtree(PIPELINE_ARTIFACT_DIR, ignore_errors=True)\n", "shutil.rmtree(PIPELINE_SEGMENT_DIR, ignore_errors=True)\n", "os.makedirs(PIPELINE_ARTIFACT_DIR)\n", "os.makedirs(PIPELINE_SEGMENT_DIR)" ] }, { "cell_type": "code", "execution_count": 74, "id": "e1cfe9df", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 5099.46it/s]\n" ] } ], "source": [ "manifest_filepath = os.path.join(PIPELINE_ARTIFACT_DIR , \"00_manifest.json\")\n", "manifest = []\n", "for audio_rev_meta in tqdm.tqdm(audio_rev_metas):\n", " transcript_fp = os.path.join(FROM_REV_DIR, audio_rev_meta[\"uid\"] + \".json\")\n", " if not os.path.exists(transcript_fp):\n", " continue\n", " manifest.append({\n", " \"id\": audio_rev_meta[\"uid\"],\n", " \"audio_filepath\": audio_rev_meta[\"filepath\"],\n", " \"transcript_filepath\": transcript_fp,\n", " \"segments_meta\": audio_rev_meta[\"segments_meta\"],\n", " })\n", "with open(manifest_filepath, \"w\") as f:\n", " json.dump(manifest, f)" ] }, { "cell_type": "code", "execution_count": null, "id": "2f53c224", "metadata": {}, "outputs": [], "source": [ "CUDA_VISIBLE_DEVICES=1 python -m suno_utils.scripts.customer.sanas.run \\\n", " --output_dir=\"/mnt/data-ssd-1/data/private/customer/level_ai/pilot/pipeline/2022_10_20\" \\\n", " --use_gpu \\\n", " --conda-env=\"mfa\" \\\n", " --conda_sh_filepath=\"/home/georg/anaconda3/etc/profile.d/conda.sh\" \\\n", " --mfa_dictionary_name=\"english_mfa\" \\\n", " --mfa_acoustic_model_name=\"english_mfa\" \\\n", " --num_cores=20 \\\n", " --manifest_file=\"/mnt/data-ssd-1/data/private/customer/level_ai/pilot/pipeline/2022_10_20/artifacts/00_manifest.json\" \\\n", " --data_delivery_dir=\"/mnt/data-ssd-1/data/private/customer/level_ai/pilot/pipeline/2022_10_20/artifacts/tmp_delivery\" \\\n", " --sample_rate=16000 \\\n", " --max_cer_thres=0.6 \\\n", " --skip_verify_boundaries\n", "\n", "# --anonymize_speakers \\" ] }, { "cell_type": "code", "execution_count": null, "id": "caa34e53", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "13f1f39c", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "b721d110", "metadata": {}, "source": [ "## QA" ] }, { "cell_type": "code", "execution_count": 140, "id": "9b4260c1", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "1.0 hours\n" ] } ], "source": [ "with open(os.path.join(ARTIFACTS_DIR, \"03_segment_meta_post_timestamp.json\")) as f:\n", " segments_manifest = json.load(f)\n", "print(round(np.sum([e[\"duration_s\"] for e in segments_manifest]) / 60 / 60, 1), \"hours\")" ] }, { "cell_type": "code", "execution_count": 141, "id": "14d34ae2", "metadata": {}, "outputs": [ { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "237\n", "She's confirming for me now, Aaliyah.\n", "she's confirming for me now aaliyah\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "139\n", "One was the chicken something. One was the all meaty. Mm-hmm. And that's for 10:45 to 2000 Stanley [inaudible] Mm-hmm. Okay.\n", "one was the chicken something one was the all meaty mm hmm and that's for ten forty five to two thousand stanley [inaudible] mm hmm okay\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "1\n", "\n", "\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "203\n", "Hi, Stephanie. This is, uh, Joe calling from the ezCater Partner Success Department. How's it going? I'm all right. How are you? I'm doing well. Um, yeah. I'm just calling because, um, it loo -- You know, looks like w -- uh, you know, you're trying to get back on the ezCater site, which --\n", "hi stephanie this is uh joe calling from the ezcater partner success department how's it going i'm all right how are you i'm doing well um yeah i'm just calling because um it loo -- you know looks like w -- uh you know you're trying to get back on the ezcater site which --\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "253\n", "And the size is the pan. Okay. Two gallon sweet teas. Sorry, three gallon sweet tea.\n", "and the size is the pan okay two gallon sweet teas sorry three gallon sweet tea\n", "----------\n" ] } ], "source": [ "# look at random ones\n", "l = list(zip(range(len(segments_manifest)), segments_manifest[:]))\n", "random.shuffle(l)\n", "for n, m in l[:5]:\n", " play_audio(m[\"uri\"])\n", " print(n)\n", " print(Tokens.from_dict(m[\"transcript\"][\"tokens\"]).text)\n", " print(Tokens.from_dict(m[\"transcript_normalized\"][\"tokens\"]).text)\n", " print(\"-\"*10)" ] }, { "cell_type": "code", "execution_count": 142, "id": "f17c413a", "metadata": {}, "outputs": [ { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "255\n", "Sorry. Family Medicine. It's located at 2450 India Hook Road in Rock Hill. Okay. And then you said it needs to be there at 11:15? Yeah.\n", "sorry family medicine it's located at twenty four fifty india hook road in rock hill okay and then you said it needs to be there at eleven fifteen yeah\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "248\n", "Tomorrow. Okay. If we could -- At 11:15. Just, uh, go over that one, that one would be [inaudible] and then we could just go ahead and, uh, confirm that one as well. Sure. I think the -- Did you say your name was Jennifer? Correct.\n", "tomorrow okay if we could -- at eleven fifteen just uh go over that one that one would be [inaudible] and then we could just go ahead and uh confirm that one as well sure i think the -- did you say your name was jennifer correct\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "125\n", "Uh, there's a phone number associated. Okay. May I have the phone number, please? 502-303-6670.\n", "uh there's a phone number associated okay may i have the phone number please five oh two three oh three six six seven zero\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "171\n", "Hello. Thank you for calling ezCater. This is Julian speaking on a recorded line. How can I help you today? Hi. My, uh, upcoming order number is Q7X, mm, ZG9.\n", "hello thank you for calling ezcater this is julian speaking on a recorded line how can i help you today hi my uh upcoming order number is q seven x mm zg nine\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "226\n", "Yeah. So we received an update that the caterer was actually on their way to deliver the food. Yeah. I know. They were on the way to deliver the food at 12:00. They shoulda been there at 12:15. No updates from there?\n", "yeah so we received an update that the caterer was actually on their way to deliver the food yeah i know they were on the way to deliver the food at twelve o'clock they shoulda been there at twelve fifteen no updates from there\n", "----------\n" ] } ], "source": [ "# look at tough numbers\n", "l = [\n", " (n, m) for n, m in enumerate(segments_manifest)\n", " if re.search(r\"[^\\s0-9][0-9]+[^\\s0-9]\", Tokens.from_dict(m[\"transcript\"][\"tokens\"]).text)\n", "# if re.search(r\"AT\\&T\", Tokens.from_dict(m[\"transcript\"][\"tokens\"]).text)\n", "]\n", "random.shuffle(l)\n", "for n, m in l[:5]:\n", " play_audio(m[\"uri\"])\n", " print(n)\n", " print(Tokens.from_dict(m[\"transcript\"][\"tokens\"]).text)\n", " print(Tokens.from_dict(m[\"transcript_normalized\"][\"tokens\"]).text)\n", " print(\"-\"*10)" ] }, { "cell_type": "code", "execution_count": 143, "id": "a0609f44", "metadata": {}, "outputs": [ { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "147\n", "Yeah, okay. Uh -- Let me try again. Yeah, now it's working. Perfect. Okay, great. And could I help -- And how -- How do --\n", "yeah okay uh -- let me try again yeah now it's working perfect okay great and could i help -- and how -- how do --\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "241\n", "And they had to -- [inaudible] Okay. Thank you, ma'am. Oh man [inaudible] You're welcome. [inaudible] Okay. [inaudible] Bye bye.\n", "and they had to -- [inaudible] okay thank you ma'am oh man [inaudible] you're welcome [inaudible] okay [inaudible] bye bye\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "104\n", "Thursday. [spanish]\n", "thursday [spanish]\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "1\n", "\n", "\n", "----------\n" ] }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "224\n", "For the number of people that they're going to be preparing -- order for. I can tell you right now, it's not gonna be ready til probably [inaudible] o'clock. Okay, that's fine. All right. So -- Okay, let's -- Go ahead, sorry? No, no, no.\n", "for the number of people that they're going to be preparing -- order for i can tell you right now it's not gonna be ready til probably [inaudible] o'clock okay that's fine all right so -- okay let's -- go ahead sorry no no no\n", "----------\n" ] } ], "source": [ "# look at high cer tagged ones\n", "l = [(n, m) for n, m in enumerate(segments_manifest) if \"high asr cer\" in m.get(\"ignore_reason\", \"\")]\n", "random.shuffle(l)\n", "for n, m in l[:5]:\n", " play_audio(m[\"uri\"])\n", " print(n)\n", " print(Tokens.from_dict(m[\"transcript\"][\"tokens\"]).text)\n", " print(Tokens.from_dict(m[\"transcript_normalized\"][\"tokens\"]).text)\n", " print(\"-\"*10)" ] }, { "cell_type": "code", "execution_count": 208, "id": "b608a979", "metadata": {}, "outputs": [], "source": [ "# # find specific one:\n", "# for m in segments_manifest:\n", "# if \"11-year\" in Tokens.from_dict(m[\"transcript\"][\"tokens\"]).text:\n", "# print(m)\n", "# break" ] }, { "cell_type": "code", "execution_count": null, "id": "9352a470", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "5f51345f", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "33ae5fab", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "32c3f559", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "ae98f9e9", "metadata": {}, "source": [ "## make deliverable" ] }, { "cell_type": "code", "execution_count": 110, "id": "b9a2d47b", "metadata": {}, "outputs": [], "source": [ "# mistakes:\n", "\n", "# norm:\n", "# 48 - j a y" ] }, { "cell_type": "code", "execution_count": 175, "id": "b1795289", "metadata": {}, "outputs": [], "source": [ "with open(os.path.join(TO_REV_DIR, f\"audio_rev_metas.json\")) as f:\n", " audio_rev_metas = json.load(f)\n", "out_meta_map = {}\n", "for m in audio_rev_metas:\n", " for mm in m[\"segments_meta\"]:\n", " if mm[\"type\"] != \"speech\":\n", " continue\n", " out_meta_map[mm[\"id\"]] = {\n", " \"id\": mm[\"original_audio_id\"],\n", " \"start_s\": mm[\"original_audio_start_s\"],\n", " \"end_s\": mm[\"original_audio_end_s\"],\n", " }\n", " \n", "with open(os.path.join(ARTIFACT_DIR, \"raw_data_manifest.json\")) as f:\n", " filepaths_info = json.load(f)\n", "original_id_map = {}\n", "for m in filepaths_info:\n", " original_id_map[m[\"uid\"]] = \".\".join(m[\"original_filename\"].split(\".\")[:-1])\n", " \n", "with open(os.path.join(PIPELINE_ARTIFACT_DIR, \"03_segment_meta_post_timestamp.json\")) as f:\n", " segments_manifest = json.load(f)" ] }, { "cell_type": "code", "execution_count": 176, "id": "a2863185", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "12\n", "1.0 hours\n" ] } ], "source": [ "def _uuid_hash(s):\n", " return str(uuid.UUID(hashlib.sha256(s.encode('utf-8')).hexdigest()[::2]))\n", "\n", "output_segments_manifest = []\n", "for m in segments_manifest:\n", " uid = m[\"id\"]\n", " tokens = m[\"transcript\"][\"tokens\"]\n", " tokens_norm = m[\"transcript_normalized\"][\"tokens\"]\n", " tokens_fixed = tokens[:]\n", " tokens_norm_fixed = tokens_norm[:]\n", "# if uid == \"218a1b6e-55e7-49f3-8d84-d0ef03ba3bf6\":\n", "# tokens_norm_fixed = [\n", "# {'value': 'string', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 0.0, 'end_s': 0.52}, \n", "# {'value': 'say', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 1.5, 'end_s': 1.56}, \n", "# {'value': 'it', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 1.61, 'end_s': 1.77}, \n", "# {'value': 'again', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 1.77, 'end_s': 2.48}, \n", "# {'value': \"it's\", 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 4.11, 'end_s': 4.3}, \n", "# {'value': 'uh', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 4.3, 'end_s': 4.55}, \n", "# {'value': 'four', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 4.55, 'end_s': 4.95}, \n", "# {'value': 'four', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 5.34, 'end_s': 5.75}, \n", "# {'value': 'six', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 6.61, 'end_s': 6.89}, \n", "# {'value': 'two', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 6.89, 'end_s': 7.22}, \n", "# {'value': 'two', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 7.56, 'end_s': 7.76}, \n", "# {'value': 'one', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 7.76, 'end_s': 8.14}, \n", "# {'value': 'seven', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 8.53, 'end_s': 8.8}, \n", "# {'value': 'zero', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 8.8, 'end_s': 9.21}, \n", "# {'value': 'four', 'type': 'text', 'speaker_id': 'Speaker 1', 'start_s': 9.21, 'end_s': 9.46}, \n", "# {'value': '[inaudible]', 'type': 'tag'},\n", "# ]\n", "# print(\"fixed:\", Tokens.from_dict(tokens_norm_fixed).text)\n", "# elif uid == \"e28846c0-9d10-4179-a7f9-c7f8ee0ae904\":\n", "# tokens_norm_fixed = []\n", "# for t in tokens_norm:\n", "# if t[\"value\"] == \"zero\":\n", "# t_fixed = copy.deepcopy(t)\n", "# t_fixed[\"value\"] = \"oh\"\n", "# tokens_norm_fixed.append(t_fixed)\n", "# else:\n", "# tokens_norm_fixed.append(t)\n", "# print(\"fixed:\", Tokens.from_dict(tokens_norm_fixed).text)\n", "# elif uid == \"4eaf1934-1108-45ae-9aee-0ad68c0411f9\":\n", "# tokens_fixed = []\n", "# for t in tokens:\n", "# if t[\"value\"] == \",but\":\n", "# t_fixed = copy.deepcopy(t)\n", "# t_fixed[\"value\"] = \"but\"\n", "# tokens_fixed.append(t_fixed)\n", "# else:\n", "# tokens_fixed.append(t)\n", "# print(\"fixed:\", Tokens.from_dict(tokens_fixed).text)\n", "# elif uid == \"61ab3fae-92a1-4e86-93a4-9d7d7f86bdaf\":\n", "# tokens_fixed = []\n", "# for t in tokens:\n", "# if t[\"value\"] == \".\":\n", "# continue\n", "# else:\n", "# tokens_fixed.append(t)\n", "# print(\"fixed:\", Tokens.from_dict(tokens_fixed).text)\n", "# elif uid in (\n", "# \"36de2fa8-dd44-4ecc-9b1b-c3f2865f511d\",\n", "# \"586ff8bf-4c26-485c-934a-6c4c955c9c8d\",\n", "# \"66597f84-32f9-454b-a8ef-82dd50b92bf6\",\n", "# \"2bffd91a-a65a-4de0-abf0-d90035301b80\",\n", "# \"baca3eb0-09ba-4fb0-bfbd-75923f3daebb\",\n", "# \"d1e92357-56ba-49c1-92e0-6d4de1df860e\",\n", "# ):\n", "# tokens_fixed = []\n", "# for t in tokens:\n", "# if t[\"value\"] == \".\":\n", "# tokens_fixed.append({\"value\": \"[inaudible]\", \"type\": \"tag\"})\n", "# else:\n", "# tokens_fixed.append(t)\n", "# print(\"fixed:\", Tokens.from_dict(tokens_fixed).text)\n", " \n", " # remove tags from normalized\n", " tokens_norm_fixed = [t for t in tokens_norm_fixed if t[\"type\"] == \"text\"]\n", "\n", " # hash speakers and make unique based on original ID\n", " tmp_tokens_fixed = []\n", " for t in tokens_fixed:\n", " tn = copy.deepcopy(t)\n", " if tn.get(\"speaker_id\") is not None:\n", " tn[\"speaker_id\"] = _uuid_hash(m[\"source_audio_id\"] + tn.get(\"speaker_id\"))\n", " tmp_tokens_fixed.append(tn)\n", " tokens_fixed = tmp_tokens_fixed[:]\n", " tmp_tokens_norm_fixed = []\n", " for t in tokens_norm_fixed:\n", " tn = copy.deepcopy(t)\n", " if tn.get(\"speaker_id\") is not None:\n", " tn[\"speaker_id\"] = _uuid_hash(m[\"source_audio_id\"] + tn.get(\"speaker_id\"))\n", " tmp_tokens_norm_fixed.append(tn)\n", " tokens_norm_fixed = tmp_tokens_norm_fixed[:]\n", " \n", " out_m = {\n", " \"id\": uid,\n", " \"uri\": f\"audio/{uid}.wav\",\n", " \"source_id\": original_id_map[m[\"source_audio_id\"]],\n", " \"source_start_s\": out_meta_map[m[\"id\"]][\"start_s\"],\n", " \"source_end_s\": out_meta_map[m[\"id\"]][\"end_s\"],\n", "# \"duration_s\": get_duration_s(m[\"uri\"]),\n", " \"transcript\": {\n", " \"tokens\": tokens_fixed,\n", " \"text\": Tokens.from_dict(tokens_fixed).text,\n", " },\n", " \"transcript_normalized\": {\n", " \"tokens\": tokens_norm_fixed,\n", " \"text\": Tokens.from_dict(tokens_norm_fixed).text,\n", " },\n", " }\n", " output_segments_manifest.append(out_m)\n", "print(len(set([m[\"source_id\"] for m in output_segments_manifest])))\n", "print(round(\n", " np.sum([m[\"source_end_s\"] - m[\"source_start_s\"] for m in output_segments_manifest]) / 60 / 60, 1\n", "), \"hours\")" ] }, { "cell_type": "code", "execution_count": 257, "id": "235eefe0", "metadata": {}, "outputs": [], "source": [ "# drift adjust: 50ms drift over 200s of original audio\n", "tmp_info = defaultdict(list)\n", "for m in output_segments_manifest:\n", " tmp_info[m[\"source_id\"]].append({\n", " \"start_s\": m[\"source_start_s\"],\n", " \"transcript\": m[\"transcript\"],\n", " \"transcript_normalized\": m[\"transcript_normalized\"],\n", " })\n", "output_fullext_manifest = []\n", "for _id, m in tmp_info.items():\n", " m = sorted(m, key=lambda x: x[\"start_s\"])\n", " tokens = []\n", " tokens_norm = []\n", " prev_end_s = 0\n", " prev_norm_end_s = 0\n", " for mm in m:\n", " adjusted_offset_s = max(0, safe_round(mm[\"start_s\"] - mm[\"start_s\"] / 200 * 0.05))\n", " tmp_tokens = []\n", " for t in mm[\"transcript\"][\"tokens\"]:\n", " tn = copy.deepcopy(t)\n", " if tn.get(\"start_s\") is not None:\n", " tn[\"start_s\"] = max(prev_end_s, safe_round(tn[\"start_s\"] + adjusted_offset_s))\n", " tn[\"end_s\"] = max(tn[\"start_s\"], safe_round(tn[\"end_s\"] + adjusted_offset_s))\n", " prev_end_s = tn[\"end_s\"]\n", " tmp_tokens.append(tn)\n", " tokens.extend(tmp_tokens)\n", " tmp_tokens_norm = []\n", " for t in mm[\"transcript_normalized\"][\"tokens\"]:\n", " tn = copy.deepcopy(t)\n", " if tn.get(\"start_s\") is not None:\n", " tn[\"start_s\"] = max(prev_norm_end_s, safe_round(tn[\"start_s\"] + adjusted_offset_s))\n", " tn[\"end_s\"] = max(tn[\"start_s\"], safe_round(tn[\"end_s\"] + adjusted_offset_s))\n", " prev_norm_end_s = tn[\"end_s\"]\n", " tmp_tokens_norm.append(tn)\n", " tokens_norm.extend(tmp_tokens_norm)\n", " \n", " tokens = Tokens.from_dict(tokens)\n", " tokens_norm = Tokens.from_dict(tokens_norm)\n", " \n", " # remove timestamps\n", "# tokens = tokens.remove_timestamps()\n", "# tokens_norm = tokens_norm.remove_timestamps()\n", " \n", " output_fullext_manifest.append({\n", " \"id\": _id,\n", " \"transcript\": {\n", " \"tokens\": tokens.to_dict(),\n", " \"text\": tokens.text.replace(\"AH914 -- J\", \"AH914J\").replace(\"customer -- Placed\", \"customer placed\"),\n", " },\n", " \"transcript_normalized\": {\n", " \"tokens\": tokens_norm.to_dict(),\n", " \"text\": tokens_norm.text,\n", " },\n", " })" ] }, { "cell_type": "code", "execution_count": 254, "id": "f066950e", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 12/12 [00:20<00:00, 1.71s/it]\n" ] } ], "source": [ "# do channel marking per speaker\n", "from scipy.io import wavfile\n", "from suno_utils.tasks.asr import transcribe\n", "speaker_channel_map = {}\n", "for m in tqdm.tqdm(output_fullext_manifest):\n", " # get words by speaker\n", " speaker_words = defaultdict(set)\n", " for mm in m[\"transcript_normalized\"][\"tokens\"]:\n", " if mm[\"type\"] == \"text\":\n", " speaker_words[mm[\"speaker_id\"]].add(mm[\"value\"])\n", " # do asr transcript by channel\n", " sr, arr = wavfile.read(os.path.join(RAW_AUDIO_DIR, m[\"id\"] + \".wav\"))\n", " audio_0 = Audio.from_array(arr[:, 0], sample_rate=sr)\n", " audio_1 = Audio.from_array(arr[:, 1], sample_rate=sr)\n", " text_0 = transcribe(audio_0)\n", " text_1 = transcribe(audio_1)\n", " token_set_0 = set(normalize_whitespace(text_0).split())\n", " token_set_1 = set(normalize_whitespace(text_1).split())\n", " # assign channel for each speaker\n", " for speaker_id, word_set in speaker_words.items():\n", " if len(word_set - token_set_0) < len(word_set - token_set_1):\n", " speaker_channel_map[speaker_id] = 0\n", " else:\n", " speaker_channel_map[speaker_id] = 1" ] }, { "cell_type": "code", "execution_count": 259, "id": "447b464d", "metadata": {}, "outputs": [], "source": [ "# add channel marker to all tokens\n", "output_segments_manifest_channeled = []\n", "for m in copy.deepcopy(output_segments_manifest):\n", " new_tokens = []\n", " for mm in copy.deepcopy(m[\"transcript\"][\"tokens\"]):\n", " if \"speaker_id\" in mm:\n", " mm[\"channel_id\"] = speaker_channel_map[mm[\"speaker_id\"]]\n", " new_tokens.append(mm)\n", " m[\"transcript\"][\"tokens\"] = new_tokens\n", " new_tokens = []\n", " for mm in copy.deepcopy(m[\"transcript_normalized\"][\"tokens\"]):\n", " if \"speaker_id\" in mm:\n", " mm[\"channel_id\"] = speaker_channel_map[mm[\"speaker_id\"]]\n", " new_tokens.append(mm)\n", " m[\"transcript_normalized\"][\"tokens\"] = new_tokens\n", " output_segments_manifest_channeled.append(m)\n", " \n", "output_fullext_manifest_channeled = []\n", "for m in copy.deepcopy(output_fullext_manifest):\n", " new_tokens = []\n", " for mm in copy.deepcopy(m[\"transcript\"][\"tokens\"]):\n", " if \"speaker_id\" in mm:\n", " mm[\"channel_id\"] = speaker_channel_map[mm[\"speaker_id\"]]\n", " new_tokens.append(mm)\n", " m[\"transcript\"][\"tokens\"] = new_tokens\n", " new_tokens = []\n", " for mm in copy.deepcopy(m[\"transcript_normalized\"][\"tokens\"]):\n", " if \"speaker_id\" in mm:\n", " mm[\"channel_id\"] = speaker_channel_map[mm[\"speaker_id\"]]\n", " new_tokens.append(mm)\n", " m[\"transcript_normalized\"][\"tokens\"] = new_tokens\n", " output_fullext_manifest_channeled.append(m)" ] }, { "cell_type": "code", "execution_count": 261, "id": "c7d1afd8", "metadata": {}, "outputs": [], "source": [ "with open(os.path.join(ARTIFACT_DIR, \"delivery_segment_metadata.jsonl\"), \"w\") as f:\n", " for m in output_segments_manifest_channeled:\n", " f.write(json.dumps(m) + \"\\n\")\n", "with open(os.path.join(ARTIFACT_DIR, \"delivery_metadata.jsonl\"), \"w\") as f:\n", " for m in output_fullext_manifest_channeled:\n", " f.write(json.dumps(m) + \"\\n\")" ] }, { "cell_type": "code", "execution_count": 409, "id": "3deb6c62", "metadata": {}, "outputs": [], "source": [ "# !head -3 /mnt/data-ssd-1/data/private/customer/level_ai/pilot/artifacts/delivery_output_metadata.jsonl" ] }, { "cell_type": "code", "execution_count": 552, "id": "7330821b", "metadata": {}, "outputs": [], "source": [ "# !cp /mnt/data-ssd-1/data/private/customer/level_ai/pilot/artifacts/delivery_output_metadata.jsonl ." ] }, { "cell_type": "code", "execution_count": 262, "id": "c5d360b0", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "REc88644a5aa901bad8e1d889c9fe3849a\n", "Thank you for calling [inaudible] 250. How may I help you? Hi, this is Eileen from ez [inaudible] on our recorded line. I was calling you regarding, um, the catering or -- AH914J. The customer placed the order, but it hasn't been confirmed yet. I want to know, um, if, if it will be confirmed. Just checking in. [inaudible] What, what department is the order through? Through ezCater. Okay, but what is it that's being catered? To which store, do you mean, or -- I -- I'm not understanding. No, we've, uh -- We've got various departments throughout the store, and I'm not sure which ones do catering. Is it, um -- Is it barbecue, or is it deli type foods? Let me check. Give me just a moment. Just, my computer is frozen. It is true Texas barbecue. Okay, let me get you to that area, ma'am. One moment, please. Thank you. Barbecue. Hi, this is Eileen from ezCater -- Hello? On our recorded line. How are you? Uh, good. Awesome. What do you need? I was calling you regarding the catering order AH914J. The customer placed this order, but it hasn't been confirmed yet. Just checking in. ENK? Is that three, ENK? Uh, 14J. Ma'am? I'm sorry, what is it? The order number is A as in apple, H as in Harry, 914J as in John. 14J? Yes. Bring that here. Bring that here. I don't have time for this, ma'am. We're busy right now. That's why we're not answering the phone. It's AX7EMK is all I have. Do you have another one for today? Is it for today or tomorrow? It's for tomorrow. Okay, I don't have that r -- We'll have to check with that later. We're busy right now. Okay.\n", "\n", "----------\n", "\n", "RE3c910ece9dcaa9d11249775e3f94d566\n", "Thank you for calling ezCater. This is Christie on a monitored and recorded line. How may I help you? Yes, I am calling from Papa John's corporate office, and I was ca -- in Louisville -- Excuse me, in Louisville, Kentucky. And I was calling to see if we could get three, uh, names removed from your mailing list? Three names removed from a mailing list? I can help with that. Mm-hmm. D -- Can you -- Okay. Hold for me? Um, I just need to -- Sure. Get the form, and then we can fill it out. Okay. Okay. Thank you for holding. Thank you for holding. Now this removal request, was it from, like, direct mailers? Um, I'm guessing. Yeah. Uh, no. I mean, we're -- What do you mean, direct mailers? Like, uh, promot -- marketing promotions coming in the regular mail. Yeah. Okay. Mm-hmm. Perfect. Then I've got the right form. That's -- I just wanted to check. Okay. All right, so I'll be ready in just a second here. Okay. And the business name? Papa John's corporate office. Oh, uh, Papa John's International. International, all right. And the person's first name? Adriana, A-D-R-I-A-N-A. And the last name? Sanchez, S-A-N-C-H-E-Z. All right. And the address the promotion was sent to? 2002 Papa John's Boulevard, Suite 100 -- Okay, hold on. Just -- Louisville, Kentucky. Oh. Just a moment. Sorry. Suite -- [inaudible] 100. Okay, thank you. Kansas City? Louisville, Kentucky. 40299. I apologize. What was the zip code? 40299. Thanks. Mm-hmm. Okay. Now do the other people, um, have the same address on there? Yeah, just -- Yeah. Mm-hmm. Okay. All right. And the -- And the same business name, Papa John's International? Correct. All right. So give me just a moment. Mm-hmm. And I'll be ready to fill out those. All right. Okay. I'm still with you. I just got to fill out some information. [inaudible] No problem. And I'm ready for the next person's first name. Uh, Valarie. V-A-L-A-R-I-E. V-A-L-A-R-I-E, okay. And the last name? Giovenco. G-I-O-V-E-N-C-O. And the address? It's the same. Oh yeah, you told me that. [inaudible] [laughter] I don't need you to repeat it. [laughter] All right, so that one's submitted, and I'll t -- Tell you when I'm ready for the next one. Okay. Thanks, you too. I'm ready for the next name. Okay. It's Amy, A-M-Y. And the last name? Twomey, I think. T-W-O-M-E-Y. T-W-O-M-E-Y? Mm-hmm. And that's the last one. Okay. Well -- So if you should have any others under that umbrella, would you just go ahead and take those off as well, if you have that option to do that? I don't have those records, I'm sorry. I'm just filling out the form and sending it off to the department that, uh, processes these. Okay. Is there any way to put a note in there for them? Uh, I don't think so. Okay. Okay. So these should be removed in 30 to 60 days. Okay. All right, thank you. You're welcome. Thank you for calling ezCater. You, you're welcome. Have a good day. You too, bye -- bye. Thanks, bye -- bye. Why is Liberty not releasing? Come on.\n", "\n", "----------\n", "\n", "RE35197206ad5ddbd4d2fc29a2ec33a640\n", "Hello, this is [inaudible] ezCater [inaudible] recorded line. How can I help you? Uh, yes, I was on the phone with someone helping me. We got disconnected. Um, I placed an order on ezCater back in November. And I am just trying to order the -- You have the option of ordering the exact same thing, but yet it's $30 more expensive. And I'm just confused as, as to -- To why that is, and, um, just looking for someone to help me. Just one second. I can definitely go ahead and can help you. By any chance, can I confirm your first and last name? Wendy Birkin. Thank you so much. Uh, do you have an order number by any chance, Wendy? Um, the order number from -- Mm-hmm. November -- Yeah, all orders. Mm-hmm. Mm-hmm, I'm looking. Submitted, accept -- Accepted. Okay. The order number is C as in cat, Q8-JAY. CQ8-JAY, right? Uh -- Okay. Yes. C as in cat. Mm-hmm. Q8, the number eight. J, A as in apple, Y. Okay, thank you so much. From [inaudible] right? Yes. Okay. Yeah, I see it here. It seems actually all those oldest prices, uh, were updated already, so let me just double check, okay? Give me one second. Mm. Okay, so -- Let me s -- Give me one second, okay? Let me place you on hold just for a moment, ma'am, so I can go ahead, and I can take a look, and see what has changed on this. Okay, thank you. You're welcome. One second. Hello, ma'am? Yes. Hi, thank you so much for holding on the line. Okay, Wendy, as what I was checking, it seems that the reason why the -- uh, why it shows more expensive on the new orders -- It's because some of the prices increased because that order was last year and, um, they just update the prices. Wow, a $30 increase. Okay, I will not be -- Ordering from there anymore. Um, I'm on a very strict budget and really need to be under $15 a person and, um, I wanted to -- Redo the same thing out of convenience, but I'm not gonna pay $30 more. So I will look for another caterer. Thank you for checking into that for me. You're welcome. And anything else that I can do for you? No, that's it. Thank you. Bye -- bye. You're welcome. Have a wonderful day. Stay safe.\n", "\n", "----------\n", "\n", "REdbea31ba84636c9e1a0f5c1f8a83982a_3iCnbAd\n", "Calling ezCater. This is Jose on a recorded line. How can I help you? Yes, hi. I, um, placed an order for pick up for, let me see, Tuesday, July -- I'm sorry, June 7th. I was wondering if there's a way that I can make a modification to my order that I can't do online. Of course. I can definitely help with that. Am I talking to Jenna Smith? Okay. That's correct. Um -- Thank you. And your order -- [inaudible] Mm-hmm. I'm sorry about that. Go ahead, ma'am. [inaudible] my order number. Yes, the order -- Okay. No, I was just gonna -- TGYE2G? Correct. Okay. Yes, I do apologize. I was just wondering if -- That's okay. Um, and make your own tacos, which is, uh, the one I was trying to modify, or seeing if we can substitute pork for fish. We don't -- We don't eat pork in this office, so I was just wondering if we could substitute the -- Pork for fish? Let me go ahead and check that for you right now. Okay. Okay, so make your own taco. Instead of pork -- Yeah. You wanted fish. Yeah, like, a white fish is fine. Got it. Yep, I'm going to, uh, call the catering partner to see if they are able to accommodate this. If they are, I'll -- Uh, do you mind staying on hold, or do you want me to call you back as soon as, um, I have this [inaudible] Yeah, I don't mind staying on hold. It's okay. Okay, thank you. I'll call you ba -- Uh, stay on hold. Mm-hmm. Thank you. Okay, yep. Yep, I know what you meant. Thank you. Thank you for calling [inaudible] Meat Packing. How can I help you? Hi, this is Jose from ezCater on our recorded line. Is there anybody there that I could talk to regarding a catering order? Sure. Let me find the manager. Just give me a few minutes. Thank you. I appreciate it. You're welcome. Thank you. Good afternoon. Thank you for calling [inaudible] Meat Packing. This is Laura speaking. How can I help you? Hi, Laura. It seems that the phone call got dropped. Yes, how can I help you today? Yes, uh, it's -- I'm Jose from ezCater on a recorded line. Is there anyone that I can talk to regarding a catering order? Yeah, that would be me. Uh, Laura, thank you. I appreciate it. Mm-hmm. Okay. All right. This is what's going on, ma'am. I have an order that is for Jenna Smith for June the 7th. Mm-hmm. And, uh, the guest wants to know if instead of doing pork as protein you can accommodate fish. White fish for the tacos, taco bar. Make your own taco thing. Oh, um, that I'm not sure. Um, give me just one moment. I'm gonna place you on a brief hold while I can -- Confirm with my chef. Yeah. Um, just because the -- We do that kind of in a large batch. That's why we have the certain options we have, because it's just easier to make your own that way when it's, like, those -- Batch kind of items. Let me confirm with him if it's possible for us to do -- Yeah. Um, pescado instead of the carnitas. Let me place -- Okay. You on a brief hold. Take your time. Thank you. Okay, thank you so much for holding. Um -- Yeah. So for the order, I just wanna confirm. I have that it's for seven guests. Is that correct? The head count is seven? Let me check out that right now. Yes, it's seven people. And -- Okay, if -- Go ahead. And they wanna do the, uh -- They wanna do fish instead of the pork on the make your own taco bar, correct? Yes. Okay. For that number of people, yes, we can accommodate it. If -- The group ends up growing to, like, over 10, we won't be able to. Uh, but for a small group like that, we can absolutely do pescado instead of the carnitas. Thank you, Lori. I really appreciate it. Let me just -- Ask you something really quick. If it goes -- Yes. Over 10, how about if it goes up to 10, you would still be able to? Yeah, if it goes up to 10, that's probably the most we'll be able to -- Okay. Accommodate, uh, to make -- Okay. That kind of modifica -- Got you. Tion. I don't think that that's gonna happen. It's just a question just in case the -- Yeah. Customer asks me, but yeah. Of course. Okay, thank you. I appreciate it. Uh, enjoy your day, ma'am. My pleasure. [laughter] You too, take care. Bye -- bye. Bye -- bye. Bye. Thank you very much for your holding time. I really appreciate it. Mm-hmm. I do apologize it took, uh, quite a while. But it's just that the catering partner had to place me on hold because -- Sure. She had to go to -- Mm-hmm. Uh, her chef. Her chef, you know, just to make sure that she was going -- Of course. To be able to accommodate. She said yes, you are able to do it for -- Okay. This amount of people, for seven people. So, yes. Uh, just in case you want to, uh, increase the head count -- If it goes over -- Mm-hmm. 10, she won't be able to. She's just able to do it up -- Okay. To 10, but I don't -- uh, just for you to, to know, to have in mind. Okay. That's good to know. I actually just decreased it by one re -- in, within -- just before I called you. So I don't anticipate it going over 10 people at all. Okay, got you. All right. Okay. Thank you so much for your -- So anything else I can help you with? Your assistance, and I -- That's it. You've been great. Thank you for your help. Yeah. Uh, just a little, quick question. Okay. Um, I'm, I'm gonna be adding the modification, and you are gonna be getting it. Uh, it's just the special instructions. I'm gonna be adding the special instructions for them to know. And you're gonna be getting the modifications soon after this phone call is over within five to 10 minutes. Is that okay with you? Okay. That's okay. Okay. No problem, as long as I get the update. Okay, thank you. Enjoy your day. Bye -- bye. Okay. All right, thank you. You too, bye -- bye.\n", "\n", "----------\n", "\n", "REb3ba8a0d63f82831697819377beec421\n", "This is [inaudible] with ezCater, and I'm on a [inaudible] recorded line. How can I assist today? Hi, yes. My name is Silvia Calle. I'm calling you here in the [inaudible] and 52nd [inaudible] Avenue. And how can I assist you today? Hi, yes. I'm calling in regard of order, um -- Let me see what order. No, oh, she didn't send me that. Uh, can I give you the confirmation number because I can't really see the order number? Go ahead. Uh, it's, uh, 36551220. 36 -- 551220. I haven't found a match for this [inaudible] to get an order. Do you know where the order is going? Yeah. It says here, uh, 20 West 64th Street. And it's for the 16th? Uh, today. No, it's for today, the 14th at 11:30. It's -- For, uh, [inaudible] workshop, 20 We -- 6. 20 West 64th Street, New York, New York 10023. And, uh, the order is -- For eight of the work box lunches? Uh, yeah, eight workers lunch boxes. And can spell your name for me, please? Silvia. S-I-L-V-I-A. Last name, Calle. C-A-L-L-E. Thank you for the verification, Silvia. I need to just verify this order is for Hailey Newnan. Yes. Um, we're -- The -- My delivery boy is downstairs. Um, the security car -- The security guard is not letting him go upstairs. And we're calling the client, and she's not answering the phone. Totally understand what you're saying. So I'm seeing this order was definitely placed for today. However, our customer modified this order, and the order is for the 16th. They did, uh, change the delivery address. No. It says here that it was only modified, the, the, the, the chips. That's all I got. It didn't say nothing about that that order was for the 16th. Yeah, it was actually modified yesterday af -- Ternoon at 4:16 p.m. and the modification was accepted last evening. Give me one second. [spanish] [spanish] No. [spanish] [inaudible] station lunch boxes [inaudible] Thursday. [spanish] [inaudible] Thursday. [spanish] [spanish] Okay. I'm sorry. I'll call my delivery boy right now. Thank you. You're welcome. So your order will be delivered, uh, on the 16th, right? Yes, ma'am. All right. Thanks for confirming, Silvia. Have a good day.\n", "\n", "----------\n", "\n", "REe4294b2857483dcffdca2e88bd557483_oJRVzWU\n", "[inaudible] recorded on the third line. How may I help you? Hi. My name is Nicole [inaudible] and I have an account with you guys. I was supposed to do a lunch tomorrow in Maples, but with the storm coming, I -- I need to cancel my lunch. So I need to cancel my order, and I'm on the road trying to get home before the storm. I was hoping you can cancel the order for me. Sure. I'll be more than happy to help. Um, that's with McAlister's Deli, correct? Yes, McAlister's Deli, please. Thank you very much. So they do have a, uh, cancellation policy. Let me just double check for you. One moment, please. Yes, they have a 24 hour, uh, cancellation policy there, so there might be a cancellation fee. Let me just call them to confirm, okay? Okay, thank you. You're most welcome. Would you like me -- If you could let them know it -- Yeah, if -- Sure. You could let them know it's 'cause of the storm. I, I live -- I like three hours, um, on the other side of the coast. Um, so I just -- I g -- I mean, I don't wanna put my life at risk crossing the, uh, Everglades. Do you know? I understand. I'm so sorry to hear about that by the way. So I will go ahead and reach out to them. If there is, then I will reach out to you. If not, then -- I will just go ahead and process it for you, okay? Okay, thank you. I'll send you a text message right away. You're welcome. Thank you. You're welcome. Thank you for your time and have a great day. Bye. Thanks for calling McAlister's Deli. To place a takeout order, please press one. To place a catering order, please press two. For all other inquiries, please press three. You have selected an invalid option. Please try your selection again. Thanks for calling McAlister's Deli. To place a takeout order, please press one. To place a catering order, please press two. For all other inquiries, please press three. Please leave your message for -- 239. Thanks for calling McAlister's Deli. To place a takeout order, please press one. To place a catering order, please press two. For all other inquiries, please press three.\n", "\n", "----------\n", "\n", "RE7c18c2feea3271c3b379eed3be53ea5b\n", "Thank you for calling ezCater. This is Monica on a recorded line. How can I help you? Good morning, Monica. Um, I placed an order this morning, but I think I put my -- My email was wrong. And I cannot connect to my, um, my account anymore. Oh, I'm so sorry about that -- I must have typed in the wr -- Yeah, I must have done an error when I typed in my email. Mm-hmm, yeah. So sorry about that. Let me help you out with that situation, okay? May I have your name, please? It's Lynda, L-Y-N-D-A. Nice to meet you, Lynda. And by any chance -- Do you have the order number? No, that's what [laughter] I don't have. I didn't receive the text or the email, so that's why I was, like, wondering what was wrong. And I tried to connect, and I must have typed in the wrong email. Uh, there's a phone number associated. Okay. May I have the phone number, please? 502-303-6670. Thank you so much. Allow me just a moment to pull up your account real quick. Just a moment. Okay. I see I do have an order scheduled for today at 10:45 a.m. with Jet's Pizza. Yes, that's correct. Mm-hmm. Yeah, and let me check here. Uh, could you please confirm your email address? It's Lynda, L-Y-N-D-A.khaoua, K-H-A-O-U-A@forvia, F-O -- R-V-I-A at, .com. Okay, thank you so much. Just a moment. Okay, yeah. I see that we have here -- Another email address. Yeah, it -- it's close, right? It's some -- I must have -- Mm-hmm. [laughter] I must have typed -- Yeah. An error. Yeah, let me -- Just a moment. Just a moment, please. Okay. Could you please confirm the last four digits of -- Um, your card on file, please? Oh, let me -- Okay. Let me go downstairs. Give me just a minute. Sorry. Yeah, of course. It's 9912. Thank you so much. Just a moment. Okay, and could you confirm some of the, uh, the details of your order that you scheduled for today? Some items? Uh, we -- It's pizza, like, pepperoni. I think I ordered five pepperoni pizzas, and then five -- I think there are, like, 20 pizzas, like, large in total. One was the chicken something. One was the all meaty. Mm-hmm. And that's for 10:45 to 2000 Stanley [inaudible] Mm-hmm. Okay. Thank you so much. Just a moment. Okay, so you would like some help right now with updating or with changing your email address, right? Yes, please. Okay. Could you please repeat it again? It's Lynda, L-Y-N-D-A.khaoua, K-H-A-O-U-A@forvia, F-O-R -- V-I-A.com. Thank you so much. Let me just -- Be a second. Okay. We have just, uh, changed your email address, and could you please confirm if you have access? Yeah, okay. Uh -- Let me try again. Yeah, now it's working. Perfect. Okay, great. And could I help -- And how -- How do -- Yeah. [laughter] Just quick, last question. How do I know when they confirm the order? So I see that you have here a -- You selected to receive a text message, uh, when an order is confirmed by caterer. So yeah, every time you place an order and -- When it is confirmed by the caterer, you will receive a text message. Okay, perfect. Mm-hmm. Well, thank you so -- And you can change -- Yeah, you can change that on your ezCater account, yeah -- If you would also -- Okay, per -- Like to receive an email. Okay, perfect. Thank you so much. Yeah, no worries. It was my pleasure. And can I help you with anything else at this moment? No, everything is fine. Thank you so much. Have a good day. Same to you. Have a great day. Bye -- bye. [inaudible]\n", "\n", "----------\n", "\n", "RE252698bf10aaa15e4cc1143242021b59\n", "Thank you for calling DoorDash drive support. Please hold while we retrieve your details. This session may be monitored or recorded for quality assurance and training purposes. Your continued participation in this session will be considered as express consent to be monitored or recorded. Please stay on the line after your call for a brief survey about your experience. Hello, it's [inaudible] with DoorDash escalated support. How can I help you today? Hello. Uh, this is J.C. with ezCater on a recorded line. May I have your name again? Sorry. Armani. Armani. Okay, Armani. I need to update a tip -- Please. Okay. Can I get the customer's order number? 1442731851. And the customer's name? Mary Murphy. Thank you. And what [inaudible] tip amount to? 30, $30. All right. So that tip amount has been now updated, $30 even. Is there anything else I can do for you? That will be all for today. Thank you so much, and have a great weekend. Bye -- bye. Thank you for calling DoorDash. Have a good day as well. Bye -- bye. Bye -- bye. Welcome to your after call survey. Thank you for taking the --\n", "\n", "----------\n", "\n", "REdb6afa7f9b90d2ebe37e1743c1ea7cea\n", "Hello. Thank you for calling ezCater. This is Julian speaking on a recorded line. How can I help you today? Hi. My, uh, upcoming order number is Q7X, mm, ZG9. I was hoping to be able to add to it. You wanted to add some food items to this order? Yes. Okay. Uh, let me go ahead and pull up that information. Hold on just a sec. Okay. I may have the pleasure to talk with, uh -- Ekoa Komalani Komo. That. R -- yeah, that right. Yep. [laughter] Ah, okay, sir. Um, so the order, it's, it's going to be delivered in 11 minutes. Supposed to be delivered in 11 minutes. Um, what items do you wanna add? Uh, I was hoping to add a salad to it. A salad? Yeah. [inaudible] A -- any, antipasta salad. Okay. Let me go ahead and double check, uh, what they have on the menu. Seems like the cutoff time has passed. Um -- Ah, okay. So let, let me see what I can do for you. Uh, salads and sides. Mm. Uh, they do have garden salad, potato chips, uh, pasta salad. Or you said, like, salad, uh, [inaudible] salad, right? Caesar salad? Uh, n -- no. They, they have a, uh, Italian salad. [inaudible] meats. It's, uh -- Um -- Any, antipasta salad. Okay. So it, it should be -- Pasta salad in that case, or macaroni salad, which is a serve for, for, um, four people. No. Nope. Not pasta. No pasta? Okay. It, it -- No, it is, uh -- [laughter] Okay. It's a salad with, uh, mixed meats on it. Turkey club salad or buffalo chicken salad? Hold on. I'll find it. Chicken Caesar salad. Antipasto salad. That's it. Okay. Uh, it says that it's serve for six people and it, it has [inaudible] and romaine. With black olives, uh, chickpeas [inaudible] They don't have, they don't have an individual salad? No. It's actually served for six people and it costs 34.99. [inaudible] It's a tray. Okay. So -- Is it going to be able to, er, ship with the, uh, the same order? W -- Yeah. Um, let me place you on a brief hold to call the restaurant, just -- See if they're gonna be able to accept this modification. Okay. 'Cause it's last minute modification. Okay? Please hold. All right. Thank you. You're welcome. Thank you for calling DiBella's Subs in Latham. We are located at 2 Wade Road. For store hours or safe and convenient ordering, place your curbside, to-go, or catering orders on our website at DiBellas.com or by using the DiBella's mobile app. Please hold for the next available DiBella's Subs employee. At DiBella's subs -- Thank you for calling DiBella's Subs in Latham. We are located at 2 Wade Road. For store hours or safe and convenient ordering, place your curbside, to-go, or catering orders on our website at DiBellas.com or by using the DiBella's mobile app. Please hold for the next available DiBella's Subs employee. At DiBella's Subs, it's all about the bread. Our dough is aged to perfection on a daily basis, scored, and seasoned by ha -- Thank you for calling DiBella's in Latham. This is Alex. How may I help you? Hello? Okay, bye. Hi? Thank you for calling DiBella's Subs in Latham. We are located at 2 Wade Road. For store hours or safe and convenient ordering, place your curbside, to-go, or catering orders on our website at DiBellas.com, or by using the DiBella's mobile app. Please hold for the next available DiBella's Subs employee. At DiBella's Subs, it's all about the bread. Thank you for calling DiBella's in Latham. This is Alex. How may I help you? Hi, Alex. This is Julian. I'm calling from ezCater. I'm speaking on a recorded line. How are you? I'm fine. How are you? Fantastic. Never better. [laughter] Hey, um, you know what? Uh, this is regards of an order number that is due in almost six minutes to be delivered, but the customer, it's, uh, trying to see -- The possibility to add one, uh, antipasto salad. I know that the cutoff time has passed. I can't add anything to the order because my driver is out and about to drop these [laughter] orders. I'm so sorry. Oh. Ah. Oh my [inaudible] Yeah. Okay. So you're gonna have to let the customer know that the order is gonna be there and she can't add anything. All right? Okay. Awesome. Thank you very much, Alex. [inaudible] here? Yeah. Oh. [inaudible] Okay. Have a wonderful day. Yes. ezCater rep? Yep. Yeah, yeah. This is Julian speaking from ezCater. Hi. Why are you calling me about an add-on five minutes before the delivery time? Ah, yeah. You as an ezCater representative need to just use common sense and tell the customer no. Yeah. Instead of interrupting my lunch rush. Okay? Absolutely, sir. And my name is Julio. I'm -- Thank you, sir. Have a great afternoon. Like wha -- Thank you so much for the waiting time, and I do apologize for the long delay. No, that's all right. Um, I already talked to Alex. Uh, from the restaurant and she really apologized 'cause the driver just left. Um -- Oh, that's fine. Couple of minutes ago. Uh-huh. That's, that's fine. So I'm really sorry. No, that's all right. Thank you. You're more than welcome. Besides that, do you have another inquiry for me? Uh, nope. I think we're good. Awesome. Thank you so much for calling ezCater. Stay safe -- [inaudible] And take care, and have a wonderful day. Thank you too. Bye. Bye bye.\n", "\n", "----------\n", "\n", "RE96f3d5bf906e69a7be50b2b286de2452_Jmra2JQ\n", "This is Stephanie. Hi, Stephanie. This is, uh, Joe calling from the ezCater Partner Success Department. How's it going? I'm all right. How are you? I'm doing well. Um, yeah. I'm just calling because, um, it loo -- You know, looks like w -- uh, you know, you're trying to get back on the ezCater site, which -- You know, is great. You know, we'd love to have you back on. Um, it's just that we need a, a food service permit, uh, in order to do that. Um, so if you're able to provide one for us, you know, we'd be, eh, be happy to, uh, uh, get you back on the site then. Okay. Um, I have -- Just, uh, called my, uh, county health sys -- health inspector and he said -- [inaudible] Missouri does not require that. Then I called the state, Missouri State Health inspe -- um, Health Department. And they said as long as I'm within the county requirements, no, I am not required to u -- have that form. There's no such thing in our county. Mm, uh, your county doesn't have those kinds of forms. No. Okay. That's what I'm saying. It, it, it's not relevant for, mm, me. Okay. If you would like, I would be more than happy to give you the county health inspector's information. And you are more than welcome to call him, or email him, or whatever you wanna do. Okay. Because he already said, you know, whenever I called him and talked to him, he said that he would be more than happy to speak to the, to you guys on, you know, my behalf. Okay. Uh, yeah. Absolutely. Uh, I'd be happy to do that. Uh, what, uh, what, uh, wi -- which, uh, number should I be, uh, calling here? Okay. Hang on one second here. I'm gonna pull that up. Oh. All right. It is -- So his name is Steve Darnell, like it says on the health inspector, um, you know, piece of -- Or the health, uh, piece of paper that I sent you guys. Mm-hmm. Anyways, he does two counties. He does Vernon County and Bates County. So, uh -- That actual number is -- [inaudible] I got t -- Um, it is 417-321-0418. I don't know if that's his mobile [laughter] one or if that's his actual office number. So, uh -- [inaudible] I, I'm pretty sure -- [inaudible] It's his office number 'cause I have both numbers. So that was 417-321-0148? Hang on one second here. Okay, what'd you say? Uh, 417-321-0148. Nope. 0418. 0418, okay. [inaudible] Yeah, you'll c -- Uh, no I asked, you know, [inaudible] for the health, um, inspector and it's, his name is Steve Darnell. Okay. Excellent. Yeah. Um, I will get right on that and as soon as we have, uh, an answer about that, we'll give you a call back. Okay? All right. Sounds great. All right. See you, Stephanie. All right. Thanks. Bye. [inaudible]\n", "\n", "----------\n", "\n", "RE52a32902502294d18452445f11ad3a6d\n", "[inaudible] so thank you for that [inaudible] Um -- Hi, [inaudible] This is [inaudible] at ezCater on a monitored and recorded line. How can I assist you today? [inaudible] Yes. I'm calling because I just received something from Giovanni's Pizza. My name is Ivana Johnson, saying that -- They issued a full refund for my order. And befo -- the lady I spoke with before you, she just told me that they were leaving Giovanni's to go drop the pizza off at 12:15 at [inaudible] house, which was at 300 Madison. Can you confirm that information? Sure. [inaudible] I'll be happy to assist you to confirm, um, the order's left already. And may I have the order number, please? Yeah, one second. Sure. I'm still here, Aaliyah. One second, okay? No problem. Yep, no problem. The order number is 2Y7UKQ. So union, kite, quick. Aaliyah, is there any other options for in-house catering? I know it has to be the day before, but is there, i -- is there anything that we can give them, or it doesn't work like that? So -- Is this for right [inaudible] All right. Um, I [inaudible] So I'm checking out the order right now. Um, [inaudible] would have to get, like, a card on file and they would process the order. But when that order would be ready is -- Eh, if they, if they start, like, if they process the order as of right now, 12:00, 12:30, and they have a card on file, and they know that know that somebody could be charged for the order that's about to be placed -- Uh-huh. It's already 12:30. They, but that's placing the order and -- For the number of people that they're going to be preparing -- order for. I can tell you right now, it's not gonna be ready til probably [inaudible] o'clock. Okay, that's fine. All right. So -- Okay, let's -- Go ahead, sorry? No, no, no. Aaliyah, I'm gonna call [inaudible] but, well, o -- one second. Let me see what she says. Ma'am, what did you s -- what do you see? All right. So I'm now pulling up the order, um, just so that you know that I'm just reviewing the information, just one moment please. Yeah. So we received an update that the caterer was actually on their way to deliver the food. Yeah. I know. They were on the way to deliver the food at 12:00. They shoulda been there at 12:15. No updates from there? Uh, well, we haven't gotten any update apart from this. That was the last update that we receive. Uh, just around, um, 12:10, that they -- Right. They were on their way. But I'll just go ahead [inaudible] call them to confirm, uh, what's happened here. Yeah. All right? So would it be okay -- Thank you so much. You're welcome. Would it be okay for me to place you on a brief hold so I can call the caterer? Yeah. That's fine. All right. Thank you so much. Please stay on the line. I cannot -- Yes? Hello. This is [inaudible] with ezCater [inaudible] Am I speaking with Chris? Hello? [inaudible] Can you hear me? Yes I can hear you now. Yes. Yeah. I'm just calling in regards to the order that you had, you have for delivery. Um, the last update received that you were on your way to deliver the order, and the order is still not [inaudible] yet. I'm here. You're there now? I'm here. I'm here. All right. I'm here. Thank you so much for confirming. Okay. Thank you. All right. Thank you so much for holding. Um, Ivani? Yes? Yeah. So I just spoke with the c -- uh, driver and he stated that he's there now at the location. They are there? Yes. That's correct. Are they at the loading dock? Because my girl's at the loading docks. Yes. He stated that he just got there. He said he just pulled up, Aaliyah. He said he pulled up? Did, did he pull up to the loadi -- loading dock? Because Aaliyah's there. Okay. I'm not sure if he already in the loading dock, but he just, I called him and he said that he's there now. Okay. He said he's there now, Aaliyah. I know you'll be able to [inaudible] And not at the -- And not at the lobby? And they're not in the lobby, right? They're at the loading dock. Or you don't know? Yeah, they're not in lobby. He st -- he doesn't say, he didn't say where he's at. But he stated that he's just, he's there now. Can you ca -- All right. Are you able to contact him or give me his number to find exactly where he is so the young lady can pick up the food? Yeah. I can call him back to know exactly where he is. Thank you so much. You're welcome. She's confirming for me now, Aaliyah. Your call has been forwarded to an automated voice messaging system. At the tone, please record your message. Hi, Chris. This is [inaudible] with ezCater. I am just want to confirm where exactly you're at, at the customer's location since they're actually waiting, and, for the order. When you get this message, you can reach out to us at 1-800-488-1803. Thank you. Have a lovely day. All right. All right, thank you so much for holding, Ivana. So I did try to call the driver again, but he's not answering. Oh. One second. And they had to -- [inaudible] Okay. Thank you, ma'am. Oh man [inaudible] You're welcome. [inaudible] Okay. [inaudible] Bye bye.\n", "\n", "----------\n", "\n", "REf1206aa842c2db6f15b2756153f7b1f8\n", "Hi. This is Akesha with ezCater on a recorded and monitored line. How can I assist you today? Oh, good morning, ma'am. My name is Jennifer Blanks and I'm calling from Sats Café in Rock Hill. I was wondering if you could, uh, confirm an order for us. If you could accept it on your end. Mm-hmm. Sure. No problem. I can certainly assist with having that done for you. May I have the order number, please? I don't have that right in front of me, but I do know that it's going to 124 Professional Park Drive. 11:15. That order is for today? I believe so. Yes, ma'am. I see there is one that's -- Just confirm that that's -- Going out at 1:15. It's going to the Carolina Pulmonary Physicians? Okay. And what time is that going out? 11:15. Yeah. Okay. So I'm sorry, that's the one. Uh, and what's that address? It's at 124 Glenwood Drive. Yes. That's what I meant to say. Yes. So that's the one I'm talking about. And if you could just accept it on your end, that would be awesome. I saw that it was already reconfirmed. Okay. We'd, we just got an automated voicemail saying that we needed to confirm it, so that's how I got you on the line. Okay. So I do see there's another order that was recently submitted. It's for -- Tomorrow. Okay. If we could -- At 11:15. Just, uh, go over that one, that one would be [inaudible] and then we could just go ahead and, uh, confirm that one as well. Sure. I think the -- Did you say your name was Jennifer? Correct. Hi, Jennifer. So I do have the order pulled up here. Would you like the order number? Uh, no. That's perfectly fine. Awesome. So I do have it pulled up here. It's another 11:15 AM -- Delivery time for tomorrow. I'll go ahead, and you said you want it be reconfir -- Sorry. Confirmed? Yep. And then if we could just go over the order while I've got you on the phone? Sure. So it consists of two world famous -- Calabash chicken tray, two mixed green salads. Okay, so just one moment. So the two calabash chicken trays, are they 40 counts or 60 counts? 60. Two 60 calabash chicken trays. Okay. And what else? Two mixed green salads. Okay. With oil and vinegar dressing. Okay. The, there's two ho -- homemade style mac and cheese. And the size is the pan. Okay. Two gallon sweet teas. Sorry, three gallon sweet tea. And one gallon unsweetened tea. Okay. All right. That's all. And where is that going to? It's going to Dr. Givens. It's, uh, Riverview Medical Medicines. Sorry. Family Medicine. It's located at 2450 India Hook Road in Rock Hill. Okay. And then you said it needs to be there at 11:15? Yeah. Tomorrow? Okay. And that date is gonna be the 8th. So [inaudible] eight 11:15. All right, now is there a delivery fee or a tip? Let me check on that. I see there's a delivery fee, a $10.00 delivery fee. Okay. Any tips? No tips. Okay. All right. Perfect so yeah, if you just wanna confirm that one for tomorrow at 11:15, we would appreciate that. Sh -- Sure. I just went ahead and accepted it for you. Is there anything else I can assist with before I go? Nope, that's it. Thank you so much. You're welcome. You have a wonderful day now. Bye! You as well. Bye bye.\n", "\n", "----------\n", "\n" ] } ], "source": [ "for m in output_fullext_manifest:\n", " print(m[\"id\"])\n", " print(m[\"transcript\"][\"text\"])\n", "# print(m[\"transcript_normalized\"][\"text\"])\n", " print()\n", " print(\"-\"*10)\n", " print()" ] }, { "cell_type": "code", "execution_count": 191, "id": "bfc24b7c", "metadata": {}, "outputs": [], "source": [ "orig_id = \"RE96f3d5bf906e69a7be50b2b286de2452_Jmra2JQ\"\n", "audio = Audio.from_file(f\"/mnt/data-ssd-1/data/private/customer/level_ai/pilot/raw_data/audio/{orig_id}.wav\")" ] }, { "cell_type": "code", "execution_count": null, "id": "60cd4e2e", "metadata": {}, "outputs": [], "source": [ "The customer -- Placed the order\n", "order AH914 -- J" ] }, { "cell_type": "code", "execution_count": 192, "id": "e08549a0", "metadata": {}, "outputs": [ { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "audio.play()" ] }, { "cell_type": "code", "execution_count": 189, "id": "05b3f946", "metadata": {}, "outputs": [ { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "audio.play(197.32, 197.51)" ] }, { "cell_type": "code", "execution_count": 181, "id": "c7dd90d3", "metadata": {}, "outputs": [], "source": [ "segment_id = \"078d75f3-5c3c-480f-bcb8-e25368a09897\"\n", "d = \"/mnt/data-ssd-1/data/private/customer/level_ai/pilot/pipeline/2022_10_20/\"\n", "audio_segment = Audio.from_file(f\"{d}segments/{segment_id}.wav\")" ] }, { "cell_type": "code", "execution_count": 182, "id": "d25ea428", "metadata": {}, "outputs": [ { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "audio_segment.play(9.38, 9.57)" ] }, { "cell_type": "code", "execution_count": null, "id": "317d212e", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "b837b702", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "dd7557f2", "metadata": {}, "source": [ "## Package it all up" ] }, { "cell_type": "markdown", "id": "906665fe", "metadata": {}, "source": [ "for segments: \n", "/mnt/data-ssd-1/data/private/customer/level_ai/pilot/artifacts/delivery_segment_metadata.jsonl\n", "with audio filepaths here: \n", "f\"/mnt/data-ssd-1/data/private/customer/level_ai/pilot/pipeline/2022_10_20/segments/{m['id']}.wav\"\n", "\n", "for full transcripts:\n", "/mnt/data-ssd-1/data/private/customer/level_ai/pilot/artifacts/delivery_metadata.jsonl\n", "with audio filepaths here: \n", "f\"/mnt/data-ssd-1/data/private/customer/level_ai/pilot/raw_data/audio/{m['id']}.wav\"" ] }, { "cell_type": "code", "execution_count": 263, "id": "206f90e8", "metadata": {}, "outputs": [], "source": [ "!mkdir -p /mnt/data-ssd-1/data/private/customer/level_ai/pilot/delivery\n", "!mkdir -p /mnt/data-ssd-1/data/private/customer/level_ai/pilot/delivery/suno_pilot\n", "!mkdir -p /mnt/data-ssd-1/data/private/customer/level_ai/pilot/delivery/suno_pilot/audio" ] }, { "cell_type": "code", "execution_count": 264, "id": "2fd542d4", "metadata": {}, "outputs": [], "source": [ "!cp /mnt/data-ssd-1/data/private/customer/level_ai/pilot/artifacts/delivery_segment_metadata.jsonl /mnt/data-ssd-1/data/private/customer/level_ai/pilot/delivery/suno_pilot/segment_metadata.jsonl\n", "!cp /mnt/data-ssd-1/data/private/customer/level_ai/pilot/artifacts/delivery_metadata.jsonl /mnt/data-ssd-1/data/private/customer/level_ai/pilot/delivery/suno_pilot/metadata.jsonl\n", "!cp /mnt/data-ssd-1/data/private/customer/level_ai/pilot/pipeline/2022_10_20/segments/* /mnt/data-ssd-1/data/private/customer/level_ai/pilot/delivery/suno_pilot/audio/" ] }, { "cell_type": "code", "execution_count": null, "id": "4005ef49", "metadata": {}, "outputs": [], "source": [ "# cd /mnt/data-ssd-1/data/private/customer/level_ai/pilot/delivery\n", "# tar -czf suno_pilot.tar.gz suno_pilot" ] }, { "cell_type": "code", "execution_count": 265, "id": "ab4a2201", "metadata": {}, "outputs": [], "source": [ "# cp /mnt/data-ssd-1/data/private/customer/level_ai/pilot/delivery/suno_pilot.tar.gz ." ] }, { "cell_type": "code", "execution_count": null, "id": "d5795880", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "9c354c09", "metadata": {}, "source": [ "{\"id\": \"RE7c18c2feea3271c3b379eed3be53ea5b\", \"transcript\": {\"tokens\": [{\"value\": \"Thank\", \"type\": \"text\", \"speaker_id\": \"cc3301c9-d42f-9928-e74e-b2614d0725de\", \"start_s\": 0.46, \"end_s\": 0.69, \"channel_id\": 0},\n", "\n", "{\"id\": \"RE252698bf10aaa15e4cc1143242021b59\", \"transcript\": {\"tokens\": [{\"value\": \"Thank\", \"type\": \"text\", \"speaker_id\": \"fa1a82f3-c3b3-9bd2-2565-18ac52cac1e3\", \"start_s\": 1.54, \"end_s\": 2.91, \"channel_id\": 1}," ] }, { "cell_type": "code", "execution_count": null, "id": "0c8ef676", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 266, "id": "66ffe357", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 267, "id": "4ec2cf35", "metadata": {}, "outputs": [ { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "uid = \"RE7c18c2feea3271c3b379eed3be53ea5b\"\n", "sr, arr = wavfile.read(os.path.join(RAW_AUDIO_DIR, uid + \".wav\"))\n", "audio_0 = Audio.from_array(arr[:, 0], sample_rate=sr)\n", "audio_1 = Audio.from_array(arr[:, 1], sample_rate=sr)\n", "audio_0.play()\n", "audio_1.play()" ] }, { "cell_type": "code", "execution_count": 268, "id": "5c212d54", "metadata": {}, "outputs": [ { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "uid = \"RE252698bf10aaa15e4cc1143242021b59\"\n", "sr, arr = wavfile.read(os.path.join(RAW_AUDIO_DIR, uid + \".wav\"))\n", "audio_0 = Audio.from_array(arr[:, 0], sample_rate=sr)\n", "audio_1 = Audio.from_array(arr[:, 1], sample_rate=sr)\n", "audio_0.play()\n", "audio_1.play()" ] }, { "cell_type": "code", "execution_count": null, "id": "a73efa95", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "90d93e36", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "55d92f14", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "242b81d9", "metadata": {}, "source": [ "## Playground" ] }, { "cell_type": "code", "execution_count": 211, "id": "e63897bc", "metadata": {}, "outputs": [], "source": [ "DELIVERY_DIR = \"/home/georg/notebooks/customers/aiera/tmp_delivery\"\n", "DELIVERY_SEGMENTS_DIR = os.path.join(DELIVERY_DIR, \"segments\")\n", "\n", "shutil.rmtree(DELIVERY_DIR, ignore_errors=True)\n", "os.makedirs(DELIVERY_DIR)\n", "os.makedirs(DELIVERY_SEGMENTS_DIR)" ] }, { "cell_type": "code", "execution_count": 210, "id": "91c17585", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "0.7 hours\n" ] } ], "source": [ "with open(os.path.join(ARTIFACTS_DIR, \"03_segment_meta_post_timestamp.json\")) as f:\n", " segments_manifest = json.load(f)\n", "print(round(np.sum([e[\"duration_s\"] for e in segments_manifest]) / 60 / 60, 1), \"hours\")" ] }, { "cell_type": "code", "execution_count": 240, "id": "83d33b88", "metadata": {}, "outputs": [], "source": [ "speakers_map = {\n", " \"Speaker 1\": \"Pete\",\n", " \"Speaker 2\": \"Bill Sibold\",\n", "}\n", "\n", "# compile long form pretty print json\n", "full_transcript_tokens = []\n", "for m in segments_manifest:\n", " full_transcript_tokens.extend(m[\"transcript\"][\"tokens\"])\n", "full_transcript_tokens = Tokens([Token.from_dict(d).remove_timestamps() for d in full_transcript_tokens])\n", "full_transcript = []\n", "for d in full_transcript_tokens.speaker_turns:\n", " text = d[\"text\"]\n", " text = re.sub(r\"-- [A-Z][a-z]\", lambda m: m.group(0).lower(), text)\n", " full_transcript.append({\n", " \"speaker_id\": speakers_map[d[\"speaker_id\"]],\n", " \"text\": text,\n", " })\n", "with open(os.path.join(DELIVERY_DIR, \"fulltext.json\"), \"w\") as f:\n", " json.dump(full_transcript, f, indent=2)\n", " \n", "# make output segments meta and write segments\n", "output_segments = []\n", "for m in segments_manifest:\n", " new_uri = os.path.join(DELIVERY_SEGMENTS_DIR, m[\"id\"] + \".wav\")\n", " shutil.copy(m[\"uri\"], new_uri)\n", " transcript = {\n", " \"tokens\": [\n", " Token.from_dict(d).add_speaker_id(speakers_map[d[\"speaker_id\"]]).as_dict() \n", " for d in m[\"transcript\"][\"tokens\"]\n", " ],\n", " \"text\": m[\"transcript\"][\"text\"],\n", " }\n", " transcript_normalized = {\n", " \"tokens\": [\n", " Token.from_dict(d).add_speaker_id(speakers_map[d[\"speaker_id\"]]).as_dict() \n", " for d in m[\"transcript_normalized\"][\"tokens\"]\n", " ],\n", " \"text\": m[\"transcript_normalized\"][\"text\"],\n", " }\n", " output_segments.append({\n", " \"id\": m[\"id\"],\n", " \"duration_s\": m[\"duration_s\"],\n", " \"uri\": os.path.join(\"segments\", m[\"id\"] + \".wav\"),\n", " \"transcript\": transcript,\n", " \"transcript_normalized\": transcript_normalized,\n", " })\n", "with open(os.path.join(DELIVERY_DIR, \"metadata.jsonl\"), \"w\") as f:\n", " for m in output_segments:\n", " f.write(json.dumps(m) + \"\\n\")" ] }, { "cell_type": "code", "execution_count": 242, "id": "826e0b06", "metadata": {}, "outputs": [], "source": [ "# !cat /home/georg/notebooks/customers/aiera/tmp_delivery/fulltext.json" ] }, { "cell_type": "code", "execution_count": null, "id": "e33e0a11", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "9b2137aa", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "ff7c5d47", "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "Python 3 (ipykernel)", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.8.10" } }, "nbformat": 4, "nbformat_minor": 5 }