{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "530aa6ae",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Populating the interactive namespace from numpy and matplotlib\n"
]
}
],
"source": [
"%pylab inline"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "5d351030",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"\""
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "eaa2cd3d",
"metadata": {},
"outputs": [],
"source": [
"import json\n",
"import pandas as pd\n",
"from suno_utils.utils.numbers import safe_round\n",
"from suno_utils.datasets.nz_parl.utils import concat_audios\n",
"from suno_utils.utils.notebook import Audio\n",
"\n",
"\n",
"MATLAB_COLORS = [\n",
" ((0.301, 0.745, 0.933), \"#4DBEEE\"),\n",
" ((0.929, 0.694, 0.125), \"#EDB120\"),\n",
" ((0.494, 0.184, 0.556), \"#7E2F8E\"),\n",
" ((0.850, 0.325, 0.098), \"#D95319\"),\n",
" ((0.466, 0.674, 0.188), \"#77AC30\"),\n",
" ((0.000, 0.447, 0.741), \"#0072BD\"),\n",
" ((0.635, 0.078, 0.184), \"#A2142F\"),\n",
"]"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "6f4a1257",
"metadata": {},
"outputs": [],
"source": [
"# aussie\n",
"# /mnt/data-ssd-1/data/gdata-mate/output\n",
"\n",
"# supreme corpus\n",
"# /mnt/data-ssd-1/data/supreme_corpus/processed_data\n",
"\n",
"# southern\n",
"# /mnt/data-ssd-1/data/WLNT/processed-data/outbound-rev/wav-files"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8f3be151",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "5e942270",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"id": "885b8abf",
"metadata": {},
"source": [
"## Aussie"
]
},
{
"cell_type": "code",
"execution_count": 145,
"id": "183a6194",
"metadata": {},
"outputs": [],
"source": [
"def get_continuous_segments(segment_metas):\n",
" segments = []\n",
" prev_uid = None\n",
" prev_speakers = set()\n",
" tmp_ts = [-1, -1]\n",
" for segment_meta in segment_metas:\n",
" start_s = segment_meta[\"source_audio_offset_s\"]\n",
" end_s = start_s + segment_meta[\"duration_s\"]\n",
" if uid == prev_uid and start_s - tmp_ts[-1] < 0.01:\n",
" # stitch\n",
" tmp_ts[-1] = end_s\n",
" if tmp_ts[0] < 0:\n",
" tmp_ts[0] = start_s\n",
" prev_speakers |= set([\n",
" t[\"speaker_id\"] for t in segment_meta[\"transcript\"][\"tokens\"] if \"speaker_id\" in t\n",
" ])\n",
" else:\n",
" if tmp_ts[0] >= 0:\n",
" segments.append((prev_uid, prev_speakers, tmp_ts[0], tmp_ts[1] - tmp_ts[0]))\n",
" tmp_ts = [start_s, end_s]\n",
" prev_speakers = set([\n",
" t[\"speaker_id\"] for t in segment_meta[\"transcript\"][\"tokens\"] if \"speaker_id\" in t\n",
" ])\n",
" prev_uid = uid\n",
" if tmp_ts[0] >= 0:\n",
" segments.append((prev_uid, prev_speakers, tmp_ts[0], tmp_ts[1] - tmp_ts[0]))\n",
" return segments"
]
},
{
"cell_type": "code",
"execution_count": 146,
"id": "50228297",
"metadata": {},
"outputs": [],
"source": [
"DATA_DIR = \"/mnt/data-ssd-1/data/gdata-mate/output/artifacts/\"\n",
"with open(DATA_DIR + \"03_segment_meta_post_timestamp.json\") as f:\n",
" segment_metas = json.load(f)\n",
"\n",
"uid = \"802920__6f1afb09-65a6-488a-be33-e094f4677d86__0__300000\"\n",
"segment_metas = [e for e in segment_metas if e[\"source_audio_id\"] == uid]\n",
"\n",
"## get longest segment (ideally is already 5 mins)\n",
"# continuous_segments = get_continuous_segments(segment_metas)\n",
"# sorted(continuous_segments, key=lambda x: x[-1])\n",
"# _, speakers, start_s, duration_s = continuous_segments[-1]\n",
"\n",
"sid_turns = []\n",
"cur_speaker = None\n",
"cur_start_s = 0\n",
"cur_end_s = 0\n",
"offset_s = 0\n",
"for e in segment_metas:\n",
" offset_s = e[\"source_audio_offset_s\"] - segment_metas[0][\"source_audio_offset_s\"]\n",
" for t in e[\"transcript\"][\"tokens\"]:\n",
" if \"speaker_id\" in t and t[\"speaker_id\"] != cur_speaker:\n",
" if cur_speaker is not None:\n",
" sid_turns.append({\n",
" \"start_s\": cur_start_s, \n",
" \"end_s\": cur_end_s, \n",
" \"name\": cur_speaker,\n",
" })\n",
" cur_start_s = safe_round(offset_s + t[\"start_s\"])\n",
" cur_end_s = safe_round(offset_s + t[\"end_s\"])\n",
" cur_speaker = t[\"speaker_id\"]\n",
" elif \"end_s\" in t:\n",
" cur_end_s = safe_round(offset_s + t[\"end_s\"])\n",
"if cur_speaker is not None:\n",
" sid_turns.append({\n",
" \"start_s\": cur_start_s, \n",
" \"end_s\": cur_end_s, \n",
" \"name\": cur_speaker,\n",
" })\n",
" \n",
"speaker_names = set([e[\"name\"] for e in sid_turns])\n",
"speaker_id_map = {name: n for n, name in enumerate(speaker_names)}\n",
"for d in sid_turns:\n",
" d[\"id\"] = speaker_id_map[d[\"name\"]]"
]
},
{
"cell_type": "code",
"execution_count": 147,
"id": "486c80e0",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"output_file: sample/tmp/sample_1.mp3 already exists and will be overwritten on build\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"audios = [Audio.from_file(e[\"uri\"]) for e in segment_metas]\n",
"audio = concat_audios(audios)\n",
"audio.convert(44_100, 2, 1).write_wav(\"sample/diarization_dreams/segments/sample_1.wav\")\n",
"audio.write_mp3(\"sample/tmp/sample_1.mp3\")\n",
"audio.play()"
]
},
{
"cell_type": "code",
"execution_count": 148,
"id": "f9e601f3",
"metadata": {},
"outputs": [],
"source": [
"segment_info = {\n",
" \"id\": \"sample_1\", \n",
" \"uri\": \"segments/sample_1.wav\", \n",
" \"duration_s\": safe_round(np.sum([e[\"duration_s\"] for e in segment_metas])), \n",
" \"speaker_annotation\": sid_turns,\n",
"}\n",
"with open(\"sample/tmp/sample_1.json\", \"w\") as f:\n",
" json.dump(segment_info, f)"
]
},
{
"cell_type": "code",
"execution_count": 149,
"id": "5d6e639a",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "iVBORw0KGgoAAAANSUhEUgAAAg0AAABfCAYAAABrwfRBAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8/fFQqAAAACXBIWXMAAAsTAAALEwEAmpwYAAAGuElEQVR4nO3dXUiU6QLA8cdxKnW1D81jBSeKSskTR1IwIiKEgiCowJsgkr2IqNugDyIKuoqimxAKitouoiIjWIKILqKb7lwR1LJOFFsstpatJ9fyOI3nYtdoXc3Hykz7/e5mnnfmeWac6O/rO++b0dfXFwAAhpMY6wUAAOODaAAAoogGACCKaAAAoogGACCKaAAAoogGACCKaAAAoogGACCKaAAAoogGACCKaAAAoogGACCKaAAAoogGACCKaAAAoogGACCKaAAAoogGACCKaAAAoogGACCKaAAAoogGACCKaAAAoogGACCKaAAAoogGACBKcqwXAABfu/r6+n8kk8nTIYQlYWL/wp0OITSlUqmtFRUVvw4cFA0AMIxkMnl61qxZiwsLC18mEom+sV7PaEmn0xnt7e2lbW1tp0MI6weOT+RaAoDPZUlhYeF/J3IwhBBCIpHoKyws7Ax/7FH5+/gXXg8AjEeJiR4M/f58nYP2gWgAAKKIBgAYBzIyMio2bNgwv/92b29vmDFjRllVVdXCEEI4f/78tH379s0a6vGtra2TFy1a9K9PWYMDIQEg0vpbLytGe44fq2bUD3Z/dnZ2urW1NburqysjNze37+rVq1OLiop6+8c3b97cGULoHM212dMAAOPE6tWrOy9fvjw9hBAuXLiQX11d3dE/dvz48YKampq5IYTw5MmT5Jo1axaUlJSUlpSUlN68efO795+npaVl8uLFi0tv376dM5L5RQMAjBNbtmzpuHTp0ozu7u6Mu3fv5ixfvvz3wbbbvn373JUrV75qbW1taW5ubikvL3/TP9bY2Dilurp64ZkzZx6tWrWqeyTz+/MEAIwTy5Yte/306dMpp06dyl+9evWQf4q4c+dOXl1d3aMQQkgmk6GgoODt8+fPMzs6OpIbN25cWFdX97CiouLNUI8fij0NADCOrF279reDBw/+s6ampmP4rf8qLy/v7Zw5c/5369at3I+Z254GAIg01EGKX9KOHTueT58+/W1lZeXra9eu5Q22zYoVK14dPXq08MCBA7+mUqnQ2dmZGUIIkyZN6rt+/frDqqqqRbm5uent27ePKDzsaQCAcWTBggW9+/fv/9t1Id534sSJn2/fvp1XXFxcumTJktKGhoas/rGpU6emb9y48Z/a2tqi8+fPTxvJ3Bl9fd/ECa4A4KM1NjY+Lisrez7W6/hSGhsbZ5aVlc0beL89DQBAFNEAAEQRDQBAlC/y7Yn1t14OeuDED6/Kw/d5Pw36mB9elf/l9lDbfaqB8wz0/rzDbTsS3+f99Fmfb6zmGImBP8ORru1jX89QjxuNz1T/PJ/63B/6tzHS7T/0no30PR1s+/fv+9Sf8dfkQ+/N5/rsfA3vz2h8VsfqdeWvf5QxJhN/Q+xpAACiiAYAIIpoAIBxICcnZ+nA+44cOVJYW1tb8KXW4IyQABCp48f5o35p7Pz1j6LPOrl79+720VzLQE7uBADD6D+501hGQ05OztLu7u6G9+/buXPnnNzc3LeHDh16duzYsZlnz54t7O3tzZg3b15PXV3do7y8vHR1dfW8rKysdFNTU86LFy8mnTx58vG5c+cK6uvrv1u6dOnvV65ceTxwLid3AoAJbPPmzS+bmprutra2tpSUlLw+fvz4zP6xzs7OZENDw73Dhw8/2bRp08Jdu3Y9e/DgQfO9e/ey79y5kx07h2gAgAmgvr4+u6KioqS4uLj0ypUrBc3Nze+uN7Fu3brfEolEKC8v7y4oKOitrKx8nZmZGYqLi18/fPhwSuwcogEAJoBt27bNr62t/fn+/fste/bs+aWnp+fd//FZWVl9IYSQmZkZJk+e/O64hEQiEVKpVPT5LRwICQCRRnKQ4pfW3d2dmDt3bm9PT0/GxYsX82fPnt37uecQDQAwDrx58yZRVFT07/7bO3bsePb++N69e3+prKxcnJ+fnyovL+/q6urK/Nxr8O0JABiGS2P/wTENAEAU0QAARBENADC8dDqd/iauovnn60wPNiYaAGB4Te3t7dMmejik0+mM9vb2aSGEpsHGfXsCAIaRSqW2trW1nW5ra1sSJvYv3OkQQlMqldo62KBvTwAAUSZyLQEAn5FoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACiiAYAIIpoAACi/B805umW7GzPuwAAAABJRU5ErkJggg==\n",
"text/plain": [
""
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
"source": [
"color_map = {}\n",
"for e in sid_turns:\n",
" if e[\"name\"] not in color_map:\n",
" color_map[e[\"name\"]] = MATLAB_COLORS[len(color_map) % len(MATLAB_COLORS)][-1]\n",
"\n",
"plt.figure(figsize=(8, 1.5))\n",
"seen_speakers = set()\n",
"for e in sid_turns:\n",
" label = e[\"name\"] if e[\"name\"] not in seen_speakers else None\n",
" seen_speakers.add(e[\"name\"])\n",
" plt.plot([e[\"start_s\"], e[\"end_s\"]], [0, 0], label=label, linewidth=5, color=color_map[e[\"name\"]])\n",
"plt.legend(bbox_to_anchor=(1.0, 0.8, 0, 0))\n",
"plt.axis(\"off\")\n",
"plt.savefig(\"sample/tmp/sample_1.png\", bbox_inches=\"tight\")\n",
"plt.show();"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c9a0bca0",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "547d1d7e",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "d262068c",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"id": "a23b1169",
"metadata": {},
"source": [
"## Southern"
]
},
{
"cell_type": "code",
"execution_count": 170,
"id": "ee26b2e1",
"metadata": {},
"outputs": [],
"source": [
"def get_continuous_segments(segment_metas):\n",
" segments = []\n",
" prev_uid = None\n",
" prev_speakers = set()\n",
" tmp_ts = [-1, -1]\n",
" for segment_meta in segment_metas:\n",
" start_s = segment_meta[\"source_audio_offset_s\"]\n",
" end_s = start_s + segment_meta[\"duration_s\"]\n",
" if uid == prev_uid and start_s - tmp_ts[-1] < 0.01:\n",
" # stitch\n",
" tmp_ts[-1] = end_s\n",
" if tmp_ts[0] < 0:\n",
" tmp_ts[0] = start_s\n",
" prev_speakers |= set([\n",
" t[\"speaker_id\"] for t in segment_meta[\"transcript\"][\"tokens\"] if \"speaker_id\" in t\n",
" ])\n",
" else:\n",
" if tmp_ts[0] >= 0:\n",
" segments.append((prev_uid, prev_speakers, tmp_ts[0], tmp_ts[1] - tmp_ts[0]))\n",
" tmp_ts = [start_s, end_s]\n",
" prev_speakers = set([\n",
" t[\"speaker_id\"] for t in segment_meta[\"transcript\"][\"tokens\"] if \"speaker_id\" in t\n",
" ])\n",
" prev_uid = uid\n",
" if tmp_ts[0] >= 0:\n",
" segments.append((prev_uid, prev_speakers, tmp_ts[0], tmp_ts[1] - tmp_ts[0]))\n",
" return segments"
]
},
{
"cell_type": "code",
"execution_count": 171,
"id": "bffd6ca5",
"metadata": {},
"outputs": [],
"source": [
"DATA_DIR = \"/mnt/data-ssd-1/data/southern-comfort/output/artifacts/\"\n",
"with open(DATA_DIR + \"03_segment_meta_post_timestamp.json\") as f:\n",
" segment_metas = json.load(f)\n",
"\n",
"\n",
"uid = \"Lenoir_City_Press_Conference_RE_-_Drug_Bust_in_Lenoir_City_2_11_2020-759010519__10-310\"\n",
"segment_metas = [e for e in segment_metas if e[\"source_audio_id\"] == uid]\n",
"\n",
"## get longest segment (ideally is already 5 mins)\n",
"# continuous_segments = get_continuous_segments(segment_metas)\n",
"# sorted(continuous_segments, key=lambda x: x[-1])\n",
"# _, speakers, start_s, duration_s = continuous_segments[-1]\n",
"\n",
"sid_turns = []\n",
"cur_speaker = None\n",
"cur_start_s = 0\n",
"cur_end_s = 0\n",
"offset_s = 0\n",
"for e in segment_metas:\n",
" offset_s = e[\"source_audio_offset_s\"] - segment_metas[0][\"source_audio_offset_s\"]\n",
" for t in e[\"transcript\"][\"tokens\"]:\n",
" if \"speaker_id\" in t and t[\"speaker_id\"] != cur_speaker:\n",
" if cur_speaker is not None:\n",
" sid_turns.append({\n",
" \"start_s\": cur_start_s, \n",
" \"end_s\": cur_end_s, \n",
" \"name\": cur_speaker,\n",
" })\n",
" cur_start_s = safe_round(offset_s + t[\"start_s\"])\n",
" cur_end_s = safe_round(offset_s + t[\"end_s\"])\n",
" cur_speaker = t[\"speaker_id\"]\n",
" elif \"end_s\" in t:\n",
" cur_end_s = safe_round(offset_s + t[\"end_s\"])\n",
"if cur_speaker is not None:\n",
" sid_turns.append({\n",
" \"start_s\": cur_start_s, \n",
" \"end_s\": cur_end_s, \n",
" \"name\": cur_speaker,\n",
" })\n",
" \n",
"speaker_names = set([e[\"name\"] for e in sid_turns])\n",
"speaker_id_map = {name: n for n, name in enumerate(speaker_names)}\n",
"for d in sid_turns:\n",
" d[\"id\"] = speaker_id_map[d[\"name\"]]"
]
},
{
"cell_type": "code",
"execution_count": 182,
"id": "6863c782",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"audios = [\n",
" Audio.from_file(\n",
" e[\"uri\"].replace(\n",
" \"/tmp/georg_pipeline_test/\", \n",
" \"/mnt/data-ssd-1/data/southern-comfort/output/\",\n",
" )\n",
" )\n",
" for e in segment_metas\n",
"]\n",
"audio = concat_audios(audios)\n",
"audio.convert(44_100, 2, 1).write_wav(\"sample/diarization_dreams/segments/sample_2.wav\")\n",
"audio.write_mp3(\"sample/tmp/sample_2.mp3\")\n",
"audio.play()"
]
},
{
"cell_type": "code",
"execution_count": 185,
"id": "1590de47",
"metadata": {},
"outputs": [],
"source": [
"segment_info = {\n",
" \"id\": \"sample_2\", \n",
" \"uri\": \"segments/sample_2.wav\", \n",
" \"duration_s\": safe_round(np.sum([e[\"duration_s\"] for e in segment_metas])), \n",
" \"speaker_annotation\": sid_turns,\n",
"}\n",
"with open(\"sample/tmp/sample_2.json\", \"w\") as f:\n",
" json.dump(segment_info, f)"
]
},
{
"cell_type": "code",
"execution_count": 189,
"id": "d17db76c",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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\n",
"text/plain": [
""
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
"source": [
"color_map = {}\n",
"for e in sid_turns:\n",
" if e[\"name\"] not in color_map:\n",
" color_map[e[\"name\"]] = MATLAB_COLORS[len(color_map) % len(MATLAB_COLORS)][-1]\n",
"\n",
"rename_map = {\n",
" 0: \"Speaker 0\"\n",
"}\n",
" \n",
"plt.figure(figsize=(8, 1.5))\n",
"seen_speakers = set()\n",
"for e in sid_turns:\n",
" label = rename_map.get(e[\"name\"], e[\"name\"]) if e[\"name\"] not in seen_speakers else None\n",
" seen_speakers.add(e[\"name\"])\n",
" plt.plot([e[\"start_s\"], e[\"end_s\"]], [0, 0], label=label, linewidth=5, color=color_map[e[\"name\"]])\n",
"plt.legend(bbox_to_anchor=(1.0, 0.8, 0, 0))\n",
"plt.axis(\"off\")\n",
"plt.savefig(\"sample/tmp/sample_2.png\", bbox_inches=\"tight\")\n",
"plt.show();"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2d7d5470",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "77f6cf5d",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "bc2a151d",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"id": "66d57465",
"metadata": {},
"source": [
"## Supreme"
]
},
{
"cell_type": "code",
"execution_count": 207,
"id": "92ab4ecf",
"metadata": {},
"outputs": [],
"source": [
"def get_continuous_segments(segment_metas):\n",
" segments = []\n",
" prev_uid = None\n",
" prev_speakers = set()\n",
" tmp_ts = [-1, -1]\n",
" for segment_meta in segment_metas:\n",
" start_s, end_s = segment_meta[\"timestamp\"]\n",
" if uid == prev_uid and start_s - tmp_ts[-1] < 0.01:\n",
" # stitch\n",
" tmp_ts[-1] = end_s\n",
" if tmp_ts[0] < 0:\n",
" tmp_ts[0] = start_s\n",
" prev_speakers |= set([\n",
" t[\"speaker_id\"] for t in segment_meta[\"transcript\"][\"tokens\"] if \"speaker_id\" in t\n",
" ])\n",
" else:\n",
" if tmp_ts[0] >= 0:\n",
" segments.append((prev_uid, prev_speakers, tmp_ts[0], tmp_ts[1] - tmp_ts[0]))\n",
" tmp_ts = [start_s, end_s]\n",
" prev_speakers = set([\n",
" t[\"speaker_id\"] for t in segment_meta[\"transcript\"][\"tokens\"] if \"speaker_id\" in t\n",
" ])\n",
" prev_uid = uid\n",
" if tmp_ts[0] >= 0:\n",
" segments.append((prev_uid, prev_speakers, tmp_ts[0], tmp_ts[1] - tmp_ts[0]))\n",
" return segments"
]
},
{
"cell_type": "code",
"execution_count": 208,
"id": "0ee53f81",
"metadata": {},
"outputs": [],
"source": [
"DATA_DIR = \"/mnt/data-ssd-1/data/supreme_corpus/processed_data/\"\n",
"with open(DATA_DIR + \"tmp_segment_meta_post_align.json\") as f:\n",
" segment_metas = json.load(f)\n",
"\n",
"## use newest uid \n",
"# meta_df = pd.read_json(\"/mnt/data-ssd-1/data/supreme_corpus/raw_data/scrape_meta.jsonl\", lines=True)\n",
"# uid = meta_df.sort_values(\"date\", ascending=False).iloc[0][\"uuid\"]\n",
"uid = \"21-511\"\n",
"segment_metas = [e for e in segment_metas if e[\"uuid\"] == uid if e[\"ignore_segment\"] == False]\n",
"\n",
"## get longest segment (trim to 3 mins)\n",
"# continuous_segments = get_continuous_segments(segment_metas)\n",
"# sorted(continuous_segments, key=lambda x: x[-1])\n",
"# _, speakers, start_s, duration_s = continuous_segments[-1]\n",
"# start_s = 1165.1\n",
"# duration_s = 2349.5\n",
"segment_start_s = 1165.1 + 2 * 5 * 60\n",
"segment_duration_s = 3 * 60\n",
"segment_metas = [\n",
" e \n",
" for e in segment_metas \n",
" if (\n",
" e[\"timestamp\"][-1] > segment_start_s and \n",
" e[\"timestamp\"][0] < segment_start_s + segment_duration_s\n",
" )\n",
"]\n",
"\n",
"sid_turns = []\n",
"cur_speaker = None\n",
"cur_start_s = 0\n",
"cur_end_s = 0\n",
"offset_s = 0\n",
"for e in segment_metas:\n",
" for t in e[\"transcript\"][\"tokens\"]:\n",
" if \"speaker_id\" in t and t[\"speaker_id\"] != cur_speaker:\n",
" if cur_speaker is not None:\n",
" sid_turns.append({\n",
" \"start_s\": cur_start_s, \n",
" \"end_s\": cur_end_s, \n",
" \"name\": cur_speaker.title(),\n",
" })\n",
" cur_start_s = safe_round(offset_s + t[\"start_s\"])\n",
" cur_end_s = safe_round(offset_s + t[\"end_s\"])\n",
" cur_speaker = t[\"speaker_id\"]\n",
" elif \"end_s\" in t:\n",
" cur_end_s = safe_round(offset_s + t[\"end_s\"])\n",
" offset_s = e[\"timestamp\"][1] - segment_metas[0][\"timestamp\"][0]\n",
"if cur_speaker is not None:\n",
" sid_turns.append({\n",
" \"start_s\": cur_start_s, \n",
" \"end_s\": cur_end_s, \n",
" \"name\": cur_speaker.title(),\n",
" })\n",
" \n",
"speaker_names = set([e[\"name\"] for e in sid_turns])\n",
"speaker_id_map = {name: n for n, name in enumerate(speaker_names)}\n",
"for d in sid_turns:\n",
" d[\"id\"] = speaker_id_map[d[\"name\"]]"
]
},
{
"cell_type": "code",
"execution_count": 209,
"id": "094190d2",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"output_file: sample/tmp/sample_3.mp3 already exists and will be overwritten on build\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"audios = [Audio.from_file(DATA_DIR + e[\"path\"]) for e in segment_metas]\n",
"audio = concat_audios(audios)\n",
"audio.convert(44_100, 2, 1).write_wav(\"sample/diarization_dreams/segments/sample_3.wav\")\n",
"audio.write_mp3(\"sample/tmp/sample_3.mp3\")\n",
"audio.play()"
]
},
{
"cell_type": "code",
"execution_count": 210,
"id": "f7f21d00",
"metadata": {},
"outputs": [],
"source": [
"segment_info = {\n",
" \"id\": \"sample_3\", \n",
" \"uri\": \"segments/sample_3.wav\", \n",
" \"duration_s\": safe_round(np.sum([e[\"timestamp\"][1] - e[\"timestamp\"][0] for e in segment_metas])), \n",
" \"speaker_annotation\": sid_turns,\n",
"}\n",
"with open(\"sample/tmp/sample_3.json\", \"w\") as f:\n",
" json.dump(segment_info, f)"
]
},
{
"cell_type": "code",
"execution_count": 211,
"id": "2b080a59",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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\n",
"text/plain": [
""
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
"source": [
"color_map = {}\n",
"for e in sid_turns:\n",
" if e[\"name\"] not in color_map:\n",
" color_map[e[\"name\"]] = MATLAB_COLORS[len(color_map) % len(MATLAB_COLORS)][-1]\n",
"\n",
"plt.figure(figsize=(8, 1.5))\n",
"seen_speakers = set()\n",
"for e in sid_turns:\n",
" label = e[\"name\"] if e[\"name\"] not in seen_speakers else None\n",
" seen_speakers.add(e[\"name\"])\n",
" plt.plot([e[\"start_s\"], e[\"end_s\"]], [0, 0], label=label, linewidth=5, color=color_map[e[\"name\"]])\n",
"plt.legend(bbox_to_anchor=(1.0, 1.03, 0, 0))\n",
"plt.axis(\"off\")\n",
"plt.savefig(\"sample/tmp/sample_3.png\", bbox_inches=\"tight\")\n",
"plt.show();"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "75ba8405",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "1e8d4e39",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "2f66d562",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "f1e0971b",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"id": "3d2be00e",
"metadata": {},
"source": [
"## Make Samples"
]
},
{
"cell_type": "markdown",
"id": "08fd64ee",
"metadata": {},
"source": [
"### Samples viewer"
]
},
{
"cell_type": "code",
"execution_count": 212,
"id": "1d9e5630",
"metadata": {},
"outputs": [],
"source": [
"import base64\n",
"\n",
"with open(\"sample/tmp/sample_1.json\") as f:\n",
" d_1 = json.load(f)\n",
"with open(\"sample/tmp/sample_2.json\") as f:\n",
" d_2 = json.load(f)\n",
"with open(\"sample/tmp/sample_3.json\") as f:\n",
" d_3 = json.load(f)\n",
"\n",
" \n",
"d_1[\"filepath\"] = \"s3://suno-static-public/datasets/diarization-dreams/sample-data/segments/sample_1.mp3\"\n",
"d_2[\"filepath\"] = \"s3://suno-static-public/datasets/diarization-dreams/sample-data/segments/sample_2.mp3\"\n",
"d_3[\"filepath\"] = \"s3://suno-static-public/datasets/diarization-dreams/sample-data/segments/sample_3.mp3\"\n",
"d_1[\"png_b64\"] = base64.b64encode(open(\"sample/tmp/sample_1.png\", \"rb\").read()).decode()\n",
"d_2[\"png_b64\"] = base64.b64encode(open(\"sample/tmp/sample_2.png\", \"rb\").read()).decode()\n",
"d_3[\"png_b64\"] = base64.b64encode(open(\"sample/tmp/sample_3.png\", \"rb\").read()).decode()\n",
"\n",
"sample_web_info = {\n",
" \"Sample\": [d_1, d_2, d_3],\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": 213,
"id": "95f0e9fc",
"metadata": {},
"outputs": [],
"source": [
"with open(\"sample/diarization_dreams/sample_manifest_web.json\", \"w\") as f:\n",
" json.dump(sample_web_info, f)"
]
},
{
"cell_type": "code",
"execution_count": 129,
"id": "a37573d6",
"metadata": {},
"outputs": [],
"source": [
"# with open(\"sample/diarization_dreams/segment_meta.jsonl\", \"w\") as f:\n",
"# for d in [d_1, d_2]:\n",
"# f.write(json.dumps(d) + \"\\n\")"
]
},
{
"cell_type": "markdown",
"id": "578cbdff",
"metadata": {},
"source": [
"### Samples Tar"
]
},
{
"cell_type": "code",
"execution_count": 214,
"id": "b269250c",
"metadata": {},
"outputs": [],
"source": [
"with open(\"sample/tmp/sample_1.json\") as f:\n",
" d_1 = json.load(f)\n",
"with open(\"sample/tmp/sample_2.json\") as f:\n",
" d_2 = json.load(f)\n",
"with open(\"sample/tmp/sample_3.json\") as f:\n",
" d_3 = json.load(f)"
]
},
{
"cell_type": "code",
"execution_count": 215,
"id": "bb9c6f54",
"metadata": {},
"outputs": [],
"source": [
"with open(\"sample/diarization_dreams/segment_meta.jsonl\", \"w\") as f:\n",
" for d in [d_1, d_2, d_3]:\n",
" f.write(json.dumps(d) + \"\\n\")"
]
},
{
"cell_type": "code",
"execution_count": 216,
"id": "70fabe4c",
"metadata": {},
"outputs": [],
"source": [
"import tarfile\n",
"\n",
"with tarfile.open(\"sample/diarization_dreams/diarization_nation_sample.tar.gz\", \"w:gz\") as tar:\n",
" for n in [1, 2, 3]:\n",
" tar.add(\n",
" f\"sample/diarization_dreams/segments/sample_{n}.wav\", \n",
" arcname=f\"segments/sample_{n}.wav\",\n",
" )\n",
" tar.add(\n",
" \"sample/diarization_dreams/segment_meta.jsonl\",\n",
" arcname=\"segment_meta.jsonl\",\n",
" )"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "317d277b",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "df043834",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "7909e44c",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"id": "be7fd263",
"metadata": {},
"source": [
"## Upload"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "24099844",
"metadata": {},
"outputs": [],
"source": [
"# s3cmd put \\\n",
"# /home/georg/notebooks/datasets/diarization_dreams/sample/diarization_dreams/diarization_nation_sample.tar.gz \\\n",
"# s3://suno-static-public/datasets/diarization-dreams/\n",
" \n",
"# s3cmd put \\\n",
"# /home/georg/notebooks/datasets/diarization_dreams/sample/diarization_dreams/sample_manifest_web.json \\\n",
"# s3://suno-static-public/datasets/diarization-dreams/sample-data/sample_manifest_web.json\n",
" \n",
"# s3cmd put \\\n",
"# /home/georg/notebooks/datasets/diarization_dreams/sample/tmp/sample_1.mp3 \\\n",
"# s3://suno-static-public/datasets/diarization-dreams/sample-data/segments/sample_1.mp3\n",
"\n",
"# s3cmd put \\\n",
"# /home/georg/notebooks/datasets/diarization_dreams/sample/tmp/sample_2.mp3 \\\n",
"# s3://suno-static-public/datasets/diarization-dreams/sample-data/segments/sample_2.mp3\n",
" \n",
"# s3cmd put \\\n",
"# /home/georg/notebooks/datasets/diarization_dreams/sample/tmp/sample_3.mp3 \\\n",
"# s3://suno-static-public/datasets/diarization-dreams/sample-data/segments/sample_3.mp3"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "75faec36",
"metadata": {},
"outputs": [],
"source": [
"# TODO: change the above to aws and use acl public"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b8f9c1bd",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "10cc3764",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"id": "00cb0ca9",
"metadata": {},
"source": [
"## Playground"
]
},
{
"cell_type": "markdown",
"id": "a966728c",
"metadata": {},
"source": [
"### Overlapping speech"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "54f7fa17",
"metadata": {},
"outputs": [],
"source": [
"# TODO: test overlapping speech segments and refine alignment using speech separation?"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "dfecd146",
"metadata": {},
"outputs": [],
"source": [
"from suno_utils.utils.conversion import Audio\n",
"from suno_utils.utils.notebook import _\n",
"from suno_utils.web.mfa import align_tokens, align_text\n",
"from suno_utils.utils.alignment import format_tokens_as_turns\n",
"\n",
"conda_sh_filepath = \"/home/georg/anaconda3/etc/profile.d/conda.sh\"\n",
"env_name = \"mfa\"\n",
"\n",
"def _time_align_norm(segment_meta):\n",
" word_aligment = align_tokens(\n",
" [DATA_DIR + segment_meta[\"path\"]], [segment_meta[\"transcript\"][\"tokens_norm\"]], \n",
" conda_sh_filepath, env_name, num_cores=1\n",
" )[0]\n",
" for t, (st, et) in zip(segment_meta[\"transcript\"][\"tokens_norm\"], word_aligment):\n",
" t[\"start_s\"] = st\n",
" t[\"end_s\"] = et"
]
},
{
"cell_type": "code",
"execution_count": 63,
"id": "7db45895",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"207\n"
]
}
],
"source": [
"for n, segment_meta in enumerate(segment_metas):\n",
" tokens = segment_meta[\"transcript\"][\"tokens\"]\n",
" n_speakers = len(set([\n",
" t[\"speaker_id\"] for t in segment_meta[\"transcript\"][\"tokens_norm\"] if \"speaker_id\" in t\n",
" ]))\n",
" if n_speakers >= 2 and n > 200:\n",
" print(n)\n",
" break\n",
"# 207 - 2 speakers interruption"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "f452f138",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"\"this policy then congress should at least have said that clearly in the statute and put it into twelve i think their argument is that it was i apologize for interrupting you but i think they suggest that there's the underlying rule that congress that you shall detain and\""
]
},
"execution_count": 4,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"segment_meta = segment_metas[207]\n",
"audio = Audio.from_file(DATA_DIR + segment_meta[\"path\"])\n",
"segment_meta[\"transcript\"][\"text_norm\"]"
]
},
{
"cell_type": "code",
"execution_count": 51,
"id": "b6f11299",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"audio.play()"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "15982456",
"metadata": {},
"outputs": [],
"source": [
"_time_align_norm(segment_meta)"
]
},
{
"cell_type": "code",
"execution_count": 68,
"id": "7421b909",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"audio.play(2, 4.06)"
]
},
{
"cell_type": "code",
"execution_count": 69,
"id": "7833818e",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"audio.play(4.06, 6)"
]
},
{
"cell_type": "code",
"execution_count": 31,
"id": "edaab1ed",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[{'type': 'text',\n",
" 'value': 'this',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 0.0,\n",
" 'end_s': 0.21},\n",
" {'type': 'text',\n",
" 'value': 'policy',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 0.24,\n",
" 'end_s': 1.04},\n",
" {'type': 'text',\n",
" 'value': 'then',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 1.08,\n",
" 'end_s': 1.22},\n",
" {'type': 'text',\n",
" 'value': 'congress',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 1.22,\n",
" 'end_s': 1.6},\n",
" {'type': 'text',\n",
" 'value': 'should',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 1.6,\n",
" 'end_s': 1.72},\n",
" {'type': 'text',\n",
" 'value': 'at',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 1.72,\n",
" 'end_s': 1.79},\n",
" {'type': 'text',\n",
" 'value': 'least',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 1.79,\n",
" 'end_s': 2.05},\n",
" {'type': 'text',\n",
" 'value': 'have',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 2.05,\n",
" 'end_s': 2.2},\n",
" {'type': 'text',\n",
" 'value': 'said',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 2.23,\n",
" 'end_s': 2.39},\n",
" {'type': 'text',\n",
" 'value': 'that',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 2.39,\n",
" 'end_s': 2.56},\n",
" {'type': 'text',\n",
" 'value': 'clearly',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 2.56,\n",
" 'end_s': 2.9},\n",
" {'type': 'text',\n",
" 'value': 'in',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 2.9,\n",
" 'end_s': 2.99},\n",
" {'type': 'text',\n",
" 'value': 'the',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 2.99,\n",
" 'end_s': 3.05},\n",
" {'type': 'text',\n",
" 'value': 'statute',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 3.05,\n",
" 'end_s': 3.48},\n",
" {'type': 'text',\n",
" 'value': 'and',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 3.48,\n",
" 'end_s': 3.57},\n",
" {'type': 'text',\n",
" 'value': 'put',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 3.57,\n",
" 'end_s': 3.66},\n",
" {'type': 'text',\n",
" 'value': 'it',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 3.66,\n",
" 'end_s': 3.74},\n",
" {'type': 'text',\n",
" 'value': 'into',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 3.74,\n",
" 'end_s': 3.91},\n",
" {'type': 'text',\n",
" 'value': 'twelve',\n",
" 'speaker_id': 'GENERAL PRELOGAR',\n",
" 'start_s': 3.91,\n",
" 'end_s': 4.06},\n",
" {'type': 'text',\n",
" 'value': 'i',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 4.06,\n",
" 'end_s': 4.09},\n",
" {'type': 'text',\n",
" 'value': 'think',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 4.09,\n",
" 'end_s': 4.21},\n",
" {'type': 'text',\n",
" 'value': 'their',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 4.21,\n",
" 'end_s': 4.3},\n",
" {'type': 'text',\n",
" 'value': 'argument',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 4.3,\n",
" 'end_s': 4.78},\n",
" {'type': 'text',\n",
" 'value': 'is',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 4.78,\n",
" 'end_s': 4.96},\n",
" {'type': 'text',\n",
" 'value': 'that',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 4.96,\n",
" 'end_s': 5.21},\n",
" {'type': 'text',\n",
" 'value': 'it',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 5.25,\n",
" 'end_s': 5.45},\n",
" {'type': 'text',\n",
" 'value': 'was',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 5.45,\n",
" 'end_s': 5.85},\n",
" {'type': 'text',\n",
" 'value': 'i',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 5.85,\n",
" 'end_s': 5.95},\n",
" {'type': 'text',\n",
" 'value': 'apologize',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 6.18,\n",
" 'end_s': 7.33},\n",
" {'type': 'text',\n",
" 'value': 'for',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 7.33,\n",
" 'end_s': 7.45},\n",
" {'type': 'text',\n",
" 'value': 'interrupting',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 7.45,\n",
" 'end_s': 8.05},\n",
" {'type': 'text',\n",
" 'value': 'you',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 8.05,\n",
" 'end_s': 8.2},\n",
" {'type': 'text',\n",
" 'value': 'but',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 8.2,\n",
" 'end_s': 8.66},\n",
" {'type': 'text',\n",
" 'value': 'i',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 8.66,\n",
" 'end_s': 8.85},\n",
" {'type': 'text',\n",
" 'value': 'think',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 9.42,\n",
" 'end_s': 9.77},\n",
" {'type': 'text',\n",
" 'value': 'they',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 9.77,\n",
" 'end_s': 9.9},\n",
" {'type': 'text',\n",
" 'value': 'suggest',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 9.9,\n",
" 'end_s': 10.67},\n",
" {'type': 'text',\n",
" 'value': 'that',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 10.67,\n",
" 'end_s': 10.98},\n",
" {'type': 'text',\n",
" 'value': \"there's\",\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 10.98,\n",
" 'end_s': 11.31},\n",
" {'type': 'text',\n",
" 'value': 'the',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 11.31,\n",
" 'end_s': 11.72},\n",
" {'type': 'text',\n",
" 'value': 'underlying',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 11.77,\n",
" 'end_s': 12.74},\n",
" {'type': 'text',\n",
" 'value': 'rule',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 12.79,\n",
" 'end_s': 13.73},\n",
" {'type': 'text',\n",
" 'value': 'that',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 14.27,\n",
" 'end_s': 14.87},\n",
" {'type': 'text',\n",
" 'value': 'congress',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 14.87,\n",
" 'end_s': 15.44},\n",
" {'type': 'text',\n",
" 'value': 'that',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 15.47,\n",
" 'end_s': 15.64},\n",
" {'type': 'text',\n",
" 'value': 'you',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 15.64,\n",
" 'end_s': 15.75},\n",
" {'type': 'text',\n",
" 'value': 'shall',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 15.75,\n",
" 'end_s': 16.24},\n",
" {'type': 'text',\n",
" 'value': 'detain',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 16.24,\n",
" 'end_s': 17.04},\n",
" {'type': 'text',\n",
" 'value': 'and',\n",
" 'speaker_id': 'JUSTICE THOMAS',\n",
" 'start_s': 18.37,\n",
" 'end_s': 19.02}]"
]
},
"execution_count": 31,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"segment_meta[\"transcript\"][\"tokens_norm\"]"
]
},
{
"cell_type": "code",
"execution_count": 47,
"id": "47cbd803",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Using cache found in /home/georg/.cache/torch/hub/JorisCos_asteroid_master\n"
]
}
],
"source": [
"# asteroid speech seperator\n",
"import asteroid\n",
"from torch import hub\n",
"import torchaudio\n",
"import torch\n",
"from suno_utils.utils.notebook import play_array\n",
"\n",
"def _arr_float_to_int(arr):\n",
" return (arr * np.iinfo(np.int16).max).clip(\n",
" np.iinfo(np.int16).min, np.iinfo(np.int16).max\n",
" ).astype(np.int16)\n",
"\n",
"def _arr_int_to_float(arr):\n",
" return arr.astype(np.float32).reshape(1, -1) / np.iinfo(np.int16).max\n",
"\n",
"# models here: https://huggingface.co/models?filter=asteroid\n",
"SPEECH_SEP_MODEL = hub.load(\n",
" \"JorisCos/asteroid\", \"conv_tasnet\", \"JorisCos/ConvTasNet_Libri2Mix_sepnoisy_16k\"\n",
")\n",
"\n",
"def separate_speech(audio):\n",
" arr = torch.from_numpy(_arr_int_to_float(audio.convert(16_000, 2, 1).array))\n",
" out = SPEECH_SEP_MODEL.separate(arr)[0].numpy()\n",
" a1 = Audio(_arr_float_to_int(out[0, :]).tobytes(), 16_000, 2, 1)\n",
" a2 = Audio(_arr_float_to_int(out[1, :]).tobytes(), 16_000, 2, 1)\n",
" return a1, a2"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d9c368d3",
"metadata": {},
"outputs": [],
"source": [
"# TODO: figure out speaker ID for each audio, then do alignment by speaker and reconstitute together"
]
},
{
"cell_type": "code",
"execution_count": 48,
"id": "353d8736",
"metadata": {},
"outputs": [],
"source": [
"a1, a2 = separate_speech(audio)"
]
},
{
"cell_type": "code",
"execution_count": 70,
"id": "45bebfba",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"a1.play()"
]
},
{
"cell_type": "code",
"execution_count": 71,
"id": "a4e1769d",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"a2.play()"
]
},
{
"cell_type": "code",
"execution_count": 52,
"id": "bbca94a8",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"\"but i think their argument is that it was a apologize for interrupting you but ah i think they suggest that there's the underlying ah rule ah that congress that you shall detain a\""
]
},
"execution_count": 52,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"a1.transcribe()"
]
},
{
"cell_type": "code",
"execution_count": 53,
"id": "d8f1d404",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"'this policy then congress should at least have said that clearly in the statute and'"
]
},
"execution_count": 53,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"a2.transcribe()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "db56e783",
"metadata": {},
"outputs": [],
"source": [
"from collections import defaultdict\n",
"from nemo.collections.asr.metrics.wer import word_error_rate\n",
"import tempfile\n",
"\n",
"def _get_cer(ref, hyp):\n",
" if ref == hyp:\n",
" return 0\n",
" cer = word_error_rate([hyp], [ref], use_cer=True)\n",
" return np.min([1.0, cer])\n",
"\n",
"def align_overlaps(audio, tokens):\n",
" \"\"\"only works for 2 speakers for now\"\"\"\n",
" audios = separate_speech(audio.convert(16_000, 2, 1))\n",
" if len(audios) < 2 :\n",
" return None\n",
" turns = format_tokens_as_turns(tokens)\n",
" audio_transcripts = [a.transcribe() for a in audios]\n",
" # group turns by speaker\n",
" tmp_turns = defaultdict(list)\n",
" for s, t in turns:\n",
" tmp_turns[s].append(t)\n",
" stitched_turns = sorted(\n",
" [(k, \" \".join(v)) for k, v in tmp_turns.items()], \n",
" key=lambda x: len(x[-1]), reverse=True,\n",
" )\n",
" if len(stitched_turns) != 2:\n",
" return None\n",
" # use cer to figure out speaker mapping and make list of audios and corresponding transcript\n",
" speaker_id_to_sep_idx = {}\n",
" for speaker_id, text in stitched_turns:\n",
" best_n, best_cer = sorted(\n",
" [(n, _get_cer(text, transcript)) for n, transcript in enumerate(audio_transcripts)],\n",
" key=lambda x: x[-1]\n",
" )[0]\n",
" if best_cer <= 0.3:\n",
" speaker_id_to_sep_idx[speaker_id] = best_n\n",
" sep_idx_list = speaker_id_to_sep_idx.values()\n",
" if len(sep_idx_list) != len(set(sep_idx_list)):\n",
" return None\n",
" align_audios = []\n",
" align_transcripts = []\n",
" for speaker_id, text in stitched_turns:\n",
" sep_idx = speaker_id_to_sep_idx[speaker_id]\n",
" align_audios.append(audios[sep_idx])\n",
" align_transcripts.append(text)\n",
" # run align on both audios\n",
" with tempfile.TemporaryDirectory() as tmp_dir:\n",
" align_audio_filepaths = []\n",
" for n, audio in enumerate(align_audios):\n",
" fp = tmp_dir + \"{}.wav\".format(n)\n",
" audio.convert(16_000, 2, 1).write_wav(fp)\n",
" align_audio_filepaths.append(fp)\n",
" word_aligments = align_text(\n",
" align_audio_filepaths, align_transcripts, \n",
" conda_sh_filepath, env_name, num_cores=1,\n",
" )\n",
" # tease back apart\n",
" sort tokens back into order and get appropriate alignments\n",
" \n",
" \n",
" \n",
" \n",
" assert(len(tokens) == len(sorted_word_alignments))\n",
" for token, (start_s, end_s) in zip(tokens, sorted_word_alignments):\n",
" token[\"start_s\"] = start_s\n",
" token[\"end_s\"] = end_s\n",
" \n",
"align_overlaps(audio, segment_meta[\"transcript\"][\"tokens_norm\"])"
]
},
{
"cell_type": "code",
"execution_count": 60,
"id": "e46c5b2e",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"defaultdict(list,\n",
" {'GENERAL PRELOGAR': ['this policy then congress should at least have said that clearly in the statute and put it into twelve'],\n",
" 'JUSTICE THOMAS': [\"i think their argument is that it was i apologize for interrupting you but i think they suggest that there's the underlying rule that congress that you shall detain and\"]})"
]
},
"execution_count": 60,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"stitched_turns"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a4438d41",
"metadata": {},
"outputs": [],
"source": [
"# TODO make proper metrics\n",
"# TODO make mfa align take audio as well"
]
},
{
"cell_type": "code",
"execution_count": 76,
"id": "16912b7e",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 78,
"id": "6c1ef6e8",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"Audio(19.1 secs, 16000 Hz, 16 bits, 1 ch)"
]
},
"execution_count": 78,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"align_audios[0]"
]
},
{
"cell_type": "code",
"execution_count": 83,
"id": "81c9a530",
"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": [
"for audio in align_audios:\n",
" audio.play()"
]
},
{
"cell_type": "code",
"execution_count": 85,
"id": "47dac7f6",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[]"
]
},
"execution_count": 85,
"metadata": {},
"output_type": "execute_result"
},
{
"data": {
"image/png": "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\n",
"text/plain": [
""
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
"source": [
"plt.plot(align_audios[0].array)"
]
},
{
"cell_type": "code",
"execution_count": 84,
"id": "12d7eb98",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[\"i think their argument is that it was i apologize for interrupting you but i think they suggest that there's the underlying rule that congress that you shall detain and\",\n",
" 'this policy then congress should at least have said that clearly in the statute and put it into twelve']"
]
},
"execution_count": 84,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"align_transcripts"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "876b11a6",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 81,
"id": "05bd5296",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[[('i', (0.0, 0.03)),\n",
" ('think', (0.03, 3.46)),\n",
" ('their', (3.46, 4.13)),\n",
" ('argument', (4.13, 4.8)),\n",
" ('is', (4.8, 4.96)),\n",
" ('that', (4.96, 5.32)),\n",
" ('it', (5.32, 5.44)),\n",
" ('was', (5.44, 5.85)),\n",
" ('i', (5.85, 5.98)),\n",
" ('apologize', (6.11, 7.33)),\n",
" ('for', (7.33, 7.45)),\n",
" ('interrupting', (7.45, 8.05)),\n",
" ('you', (8.05, 8.2)),\n",
" ('but', (8.2, 8.6)),\n",
" ('i', (8.6, 8.99)),\n",
" ('think', (9.38, 9.76)),\n",
" ('they', (9.76, 9.9)),\n",
" ('suggest', (9.9, 10.66)),\n",
" ('that', (10.66, 10.97)),\n",
" (\"there's\", (10.97, 11.29)),\n",
" ('the', (11.29, 11.77)),\n",
" ('underlying', (11.77, 12.78)),\n",
" ('rule', (12.78, 12.88)),\n",
" ('that', (12.88, 13.2)),\n",
" ('congress', (13.2, 13.93)),\n",
" ('that', (14.17, 14.95)),\n",
" ('you', (14.95, 15.28)),\n",
" ('shall', (15.28, 15.48)),\n",
" ('detain', (15.48, 16.03)),\n",
" ('and', (16.03, 17.18))],\n",
" [('this', (0.0, 0.23)),\n",
" ('policy', (0.23, 1.02)),\n",
" ('then', (1.06, 1.22)),\n",
" ('congress', (1.22, 1.6)),\n",
" ('should', (1.6, 1.72)),\n",
" ('at', (1.72, 1.79)),\n",
" ('least', (1.79, 2.05)),\n",
" ('have', (2.05, 2.2)),\n",
" ('said', (2.2, 2.39)),\n",
" ('that', (2.39, 2.56)),\n",
" ('clearly', (2.56, 2.9)),\n",
" ('in', (2.9, 2.99)),\n",
" ('the', (2.99, 3.05)),\n",
" ('statute', (3.05, 3.49)),\n",
" ('and', (3.49, 3.66)),\n",
" ('put', (3.66, 3.82)),\n",
" ('it', (3.82, 4.01)),\n",
" ('into', (4.01, 4.75)),\n",
" ('twelve', (11.76, 12.58))]]"
]
},
"execution_count": 81,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"word_aligments"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "77d3b4ba",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "41574b53",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 105,
"id": "ae61339f",
"metadata": {},
"outputs": [],
"source": [
"## import sys\n",
"## sys.path.insert(0, '/home/georg/tools/denoiser')\n",
"\n",
"# # facebook models\n",
"# from denoiser import pretrained\n",
"# class Args():\n",
"# def __init__(self, model_path=None, dns48=None, dns64=None, master64=None, valentini_nc=None):\n",
"# self.model_path = model_path\n",
"# self.dns48 = dns48\n",
"# self.dns64 = dns64\n",
"# self.master64 = master64\n",
"# self.valentini_nc = valentini_nc\n",
"# fb_models = {\n",
"# \"dns48\": pretrained.get_model(Args(dns48=True)),\n",
"# \"dns64\": pretrained.get_model(Args(dns64=True)),\n",
"# \"master64\": pretrained.get_model(Args(master64=True)),\n",
"# \"valentini_nc\": pretrained.get_model(Args(valentini_nc=True)),\n",
"# }\n",
"\n",
"# def remove_noise(audio, model_key=\"master64\", dry=0.0):\n",
"# \"\"\"using facebook model\"\"\"\n",
"# arr = _arr_int_to_float(audio.array)\n",
"# arr_enh = fb_models[model_key](torch.from_numpy(arr).view(1, -1))\n",
"# arr_enh = (arr_enh / max(arr_enh.abs().max().item(), 1)).cpu().detach().numpy()[0]\n",
"# arr_enh = dry * arr + (1 - dry) * arr_enh\n",
"# audio_enh = Audio(_arr_float_to_int(arr_enh[0, :]).tobytes(), 16_000, 2, 1)\n",
"# return audio_enh"
]
},
{
"cell_type": "code",
"execution_count": 102,
"id": "c419ab84",
"metadata": {},
"outputs": [],
"source": [
"adn1 = remove_noise(a1, model_key=\"master64\")"
]
},
{
"cell_type": "code",
"execution_count": 88,
"id": "38ba00ab",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"a1.play()"
]
},
{
"cell_type": "code",
"execution_count": 103,
"id": "d48e6a5e",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"adn1.play()"
]
},
{
"cell_type": "code",
"execution_count": 107,
"id": "d0da93af",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[]"
]
},
"execution_count": 107,
"metadata": {},
"output_type": "execute_result"
},
{
"data": {
"image/png": "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\n",
"text/plain": [
""
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
"source": [
"plt.plot(a1.array)"
]
},
{
"cell_type": "code",
"execution_count": 108,
"id": "e8dbde68",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[]"
]
},
"execution_count": 108,
"metadata": {},
"output_type": "execute_result"
},
{
"data": {
"image/png": "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\n",
"text/plain": [
""
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
"source": [
"plt.plot(a2.array)"
]
},
{
"cell_type": "code",
"execution_count": 106,
"id": "c431d9ba",
"metadata": {},
"outputs": [
{
"ename": "ImportError",
"evalue": "cannot import name 'load_pretrained' from 'demucs.separate' (/home/georg/venvs/ml/lib/python3.8/site-packages/demucs/separate.py)",
"output_type": "error",
"traceback": [
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[0;31mImportError\u001b[0m Traceback (most recent call last)",
"\u001b[0;32m/tmp/ipykernel_3813383/1859927051.py\u001b[0m in \u001b[0;36m\u001b[0;34m\u001b[0m\n\u001b[1;32m 1\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mtorchaudio\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0;32mfrom\u001b[0m \u001b[0mdemucs\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mseparate\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mload_pretrained\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mload_track\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mapply_model\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
"\u001b[0;31mImportError\u001b[0m: cannot import name 'load_pretrained' from 'demucs.separate' (/home/georg/venvs/ml/lib/python3.8/site-packages/demucs/separate.py)"
]
}
],
"source": [
"import torchaudio\n",
"from demucs.separate import load_pretrained, load_track, apply_model"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "875f4079",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "90858e6b",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"id": "4b123a60",
"metadata": {},
"source": [
"### plotting"
]
},
{
"cell_type": "code",
"execution_count": 120,
"id": "ffb51f3e",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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\n",
"text/plain": [
""
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
"source": [
"a = [(0.1, 0.3, \"speaker a\"), (0.3, 0.5, \"speaker b\"), (0.5, 1.0, \"speaker a\")]\n",
"\n",
"color_map = {}\n",
"for _, _, speaker in a:\n",
" if speaker not in color_map:\n",
" color_map[speaker] = MATLAB_COLORS[len(color_map) % len(MATLAB_COLORS)][-1]\n",
"\n",
"plt.figure(figsize=(6,2))\n",
"seen_speakers = set()\n",
"for start_s, end_s, speaker in a:\n",
" label = speaker if speaker not in seen_speakers else None\n",
" seen_speakers.add(speaker)\n",
" plt.plot([start_s, end_s], [0, 0], label=label, linewidth=5, color=color_map[speaker])\n",
"plt.legend(bbox_to_anchor=(1.0, 1.1, 0, 0))\n",
"plt.axis('off')\n",
"plt.show();"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "dd4572f0",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "9cda1ad9",
"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
}