{ "cells": [ { "cell_type": "code", "execution_count": 1, "id": "d31d8850", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:50:19.767970Z", "start_time": "2023-11-05T18:50:19.766422Z" } }, "outputs": [], "source": [ "import os\n", "os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"\"" ] }, { "cell_type": "code", "execution_count": 2, "id": "7cd4c0dd", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:50:28.635530Z", "start_time": "2023-11-05T18:50:19.950751Z" } }, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "Warning : `load_model` does not return WordVectorModel or SupervisedModel any more, but a `FastText` object which is very similar.\n" ] } ], "source": [ "import tqdm\n", "import math\n", "import torch\n", "import random\n", "import funcy\n", "import copy\n", "import gc\n", "import re\n", "import json\n", "import tempfile\n", "import collections\n", "import pandas as pd\n", "import numpy as np\n", "import fasttext\n", "from joblib import Parallel, delayed\n", "import matplotlib.pyplot as plt\n", "\n", "from suno_utils.audio import Audio\n", "from suno_utils.utils.text import (\n", " write_jsonl,\n", " read_jsonl,\n", " write_json,\n", " read_json,\n", " normalize_whitespace,\n", ")\n", "from suno_utils.utils.s3 import read_from_s3, check_s3_file_exists, open_from_s3\n", "from suno_utils.utils.tokenizers import tokenize\n", "from suno_utils.harvest.youtube.constants.text_lang import BASE_TO_FASTTEXT_REMAP\n", "\n", "LANG_ID_MODEL_FP = \"s3://suno-data/georg/trained_models/chirp_v1/lid.176.bin\"\n", "text_lang_model = read_from_s3(LANG_ID_MODEL_FP, read_f=fasttext.load_model)\n", "\n", "def _get_text_lang(text):\n", " \"\"\"get probability of input language for text\"\"\"\n", " text = text.replace(\"’\", \"'\").lower()\n", " text = re.sub(r\"\\[.+?\\]\", \" \", text)\n", " text = normalize_whitespace(text)\n", " out = text_lang_model.predict(text, k=1)\n", " lang_str = out[0][0]\n", " p_lang = out[1][0]\n", " lang = lang_str.split(\"__\")[-1]\n", " lang = BASE_TO_FASTTEXT_REMAP.get(lang, lang)\n", " # if p_lang >= 0.8:\n", " # return lang\n", " return p_lang, lang\n", "\n", "\n", "TMP_DIR = os.path.join(os.getcwd(), \"hoot\")\n", "os.makedirs(TMP_DIR, exist_ok=True)" ] }, { "cell_type": "markdown", "id": "d9d50793", "metadata": {}, "source": [ "## genius_hq" ] }, { "cell_type": "code", "execution_count": 3, "id": "8f428b8f", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:50:28.797801Z", "start_time": "2023-11-05T18:50:28.637465Z" } }, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "/home/tony/anaconda3/envs/suno_env/lib/python3.10/site-packages/Bio/pairwise2.py:278: BiopythonDeprecationWarning: Bio.pairwise2 has been deprecated, and we intend to remove it in a future release of Biopython. As an alternative, please consider using Bio.Align.PairwiseAligner as a replacement, and contact the Biopython developers if you still need the Bio.pairwise2 module.\n", " warnings.warn(\n" ] } ], "source": [ "from Bio import pairwise2\n", "\n", "from suno_utils.utils.metrics import get_cer\n", "from suno_utils.tasks.hoot import (\n", " parse_lyrics,\n", " legacy_parse_lyrics,\n", " EMBEDDING_RATE as HOOT_EMBEDDING_RATE,\n", ")\n", "from suno_utils.utils.lyrics import remove_speakers" ] }, { "cell_type": "code", "execution_count": 4, "id": "54a969c0", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:50:28.800738Z", "start_time": "2023-11-05T18:50:28.799148Z" } }, "outputs": [], "source": [ "# base_metas = read_from_s3(\"s3://suno-data/datasets/bundles/v1/genius_hq/metas.jsonl\", read_f=read_jsonl)\n", "# raw_lyrics_map = {m[\"id\"]: m[\"lyrics\"] for m in base_metas}\n", "# del base_metas\n", "# gc.collect();" ] }, { "cell_type": "code", "execution_count": 5, "id": "2cd505d5", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:52:28.514135Z", "start_time": "2023-11-05T18:50:28.802146Z" } }, "outputs": [], "source": [ "base_metas = read_from_s3(\n", " \"s3://suno-data/datasets/bundles/v1/genius_hq/metas_plus.jsonl\", read_f=read_jsonl\n", ")\n", "# TODO: original IDs are already unique here, but did we resolve correctly by highest genius views?\n", "# could be a way to filter translations that don't contain the word 'translation'" ] }, { "cell_type": "code", "execution_count": 6, "id": "6ac295a0", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:52:30.590197Z", "start_time": "2023-11-05T18:52:28.515980Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "2090009 clips\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ "100%|██████████████████████████████████████████████████████████████████████████████████████████████████| 2090009/2090009 [00:02<00:00, 1011871.25it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "3150 filtered clips\n", "192 hours\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ "\n" ] } ], "source": [ "print(len(base_metas), \"clips\")\n", "metas = []\n", "seen = set()\n", "for m in tqdm.tqdm(base_metas):\n", " # guessed_lang = _get_text_lang(m[\"lyrics\"])[1]\n", " if (\n", " m[\"genius_views\"] < 10\n", " or m[\"youtube_views\"] < 100\n", " or m[\"lang\"] is None\n", " # or m[\"lang\"].lower()[:2] != \"en\"\n", " or not (\"translation\" in m[\"genius_slug\"].lower())\n", " or m[\"duration_s\"] < 2 * 60\n", " or m[\"duration_s\"] > 8 * 60\n", " or len(m[\"lyrics\"]) < 256\n", " or m[\"id\"] in seen\n", " # or len(m[\"lyrics\"]) > 6144\n", " # or _get_text_lang(m[\"lyrics\"])[1] != \"en\"\n", " ):\n", " continue\n", " seen.add(m[\"id\"])\n", " new_m = {\n", " \"id\": m[\"id\"],\n", " \"lyrics\": m[\"lyrics\"].strip(),\n", " \"duration_s\": m[\"duration_s\"],\n", " \"lang\": m[\"lang\"],\n", " # \"guess_lang\": guessed_lang,\n", " \"audio_filepath\": m[\"audio_filepath\"],\n", " \"original_id\": m[\"original_id\"],\n", " \"genius_slug\": m[\"genius_slug\"],\n", " \"genius_views\": m[\"genius_views\"]\n", " }\n", " metas.append(new_m)\n", "print(len(metas), \"filtered clips\")\n", "print(round(sum(m[\"duration_s\"] for m in metas) / 60 / 60), \"hours\")\n", "metas_map = {m[\"id\"]: m for m in metas}\n", "# 2090009 clips\n", "# 858539 filtered clips\n", "# 52575 hours of english" ] }, { "cell_type": "code", "execution_count": 7, "id": "b828bbae", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:45:56.235249Z", "start_time": "2023-11-05T18:45:56.232949Z" } }, "outputs": [], "source": [ "# 2090009 clips\n", "# 1597044 filtered clips\n", "# 96363 hours" ] }, { "cell_type": "code", "execution_count": 8, "id": "29cb5d64", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:45:56.282681Z", "start_time": "2023-11-05T18:45:56.237099Z" } }, "outputs": [], "source": [ "# del base_metas, metas\n", "# gc.collect()" ] }, { "cell_type": "code", "execution_count": 9, "id": "aa66a026", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:45:57.531751Z", "start_time": "2023-11-05T18:45:56.284468Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Counter({'en': 984342, 'es': 94272, 'fr': 67405, 'pt': 66380, 'de': 57060, 'pl': 49185, 'ru': 46131, 'it': 40246, 'tr': 22514, 'ko': 19968, 'ja': 15112, 'iw': 10550, 'nl': 10125, 'ro': 7163, 'el': 7116, 'da': 6688, 'sv': 6429, 'ar': 6012, 'hr': 5181, 'sr': 5019, 'no': 4464, 'vi': 4152, 'fi': 4137, 'cs': 3939, 'romanization': 3938, 'tl': 3628, 'id': 3106, 'az': 2992, 'zh-Hant': 2952, 'hu': 2910, 'fa': 2635, 'bs': 2611, 'sk': 2360, 'sq': 1592, 'bg': 1464, 'uk': 1226, 'et': 1203, 'ms': 1190, 'zh': 1185, 'ca': 880, 'lv': 840, 'hi': 827, 'th': 779, 'lt': 769, 'sw': 690, 'is': 592, 'ne': 580, 'zu': 551, 'eu': 534, 'af': 497, 'sl': 468, 'mk': 455, 'co': 410, 'ht': 403, 'gl': 390, 'sco': 340, 'la': 285, 'xh': 257, 'ka': 218, 'bn': 215, 'ta': 199, 'si': 194, 'ln': 176, 'eo': 161, 'yo': 151, 'hy': 147, 'sn': 141, 'rw': 138, 'ga': 135, 'ceb': 119, 'cy': 116, 'my': 116, 'mn': 107, 'wo': 106, 'pa': 89, 'kk': 88, 'sa': 87, 'st': 83, 'jw': 80, 'uz': 78, 'war': 74, 'lb': 74, 'ak': 74, 'mi': 70, 'be': 67, 'gd': 66, 'rm': 60, 'lg': 58, 'fo': 55, 'ig': 52, 'so': 51, 'aa': 49, 'tn': 47, 'oc': 42, 'mg': 41, 'ky': 40, 'te': 39, 'sm': 39, 'haw': 38, 'ia': 34, 'ts': 33, 'tk': 32, 'ha': 32, 'am': 31, 've': 31, 'ny': 30, 'su': 29, 'br': 27, 'mt': 26, 'qu': 26, 'ie': 24, 'ur': 23, 'bh': 23, 'tt': 22, 'ku': 21, 'om': 21, 'tlh': 21, 'kha': 20, 'bi': 19, 'rn': 18, 'gn': 17, 'to': 17, 'ml': 17, 'km': 16, 'kl': 16, 'ss': 15, 'crs': 15, 'mfe': 14, 'fj': 13, 'fy': 12, 'mr': 11, 'vo': 11, 'tg': 8, 'gv': 8, 'as': 6, 'yi': 5, 'ps': 5, 'ba': 5, 'kn': 5, 'hmn': 4, 'nso': 3, 'ks': 3, 'na': 3, 'syr': 3, 'jp': 2, 'ay': 2, 'gu': 1, 'ab': 1, 'iu': 1, 'nr': 1, 'sd': 1, 'ik': 1})\n" ] } ], "source": [ "from collections import Counter\n", "c = Counter([m[\"lang\"] for m in metas])\n", "print(c)\n", "metas_map = {m[\"id\"]: m for m in metas if c[m[\"lang\"]] > 1000}" ] }, { "cell_type": "code", "execution_count": 10, "id": "2b486253", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:06.957945Z", "start_time": "2023-11-05T18:45:57.533793Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "1593894 filtered clips\n", "en 984342 60057 hours\n", "es 94272 5647 hours\n", "fr 67405 3975 hours\n", "pt 66380 3980 hours\n", "de 57060 3234 hours\n", "pl 49185 2869 hours\n", "ru 46131 2505 hours\n", "it 40246 2320 hours\n", "tr 22514 1360 hours\n", "ko 19968 1223 hours\n", "ja 15112 1014 hours\n", "iw 10550 643 hours\n", "nl 10125 570 hours\n", "ro 7163 403 hours\n", "el 7116 426 hours\n", "da 6688 392 hours\n", "sv 6429 366 hours\n", "ar 6012 379 hours\n", "hr 5181 303 hours\n", "sr 5019 283 hours\n", "no 4464 252 hours\n", "vi 4152 297 hours\n", "fi 4137 246 hours\n", "cs 3939 223 hours\n", "romanization 3938 250 hours\n", "tl 3628 258 hours\n", "id 3106 212 hours\n", "az 2992 191 hours\n", "zh-Hant 2952 204 hours\n", "hu 2910 167 hours\n", "fa 2635 175 hours\n", "bs 2611 151 hours\n", "sk 2360 137 hours\n", "sq 1592 93 hours\n", "bg 1464 81 hours\n", "uk 1226 72 hours\n", "et 1203 73 hours\n", "ms 1190 83 hours\n", "zh 1185 78 hours\n" ] } ], "source": [ "print(len(metas), \"filtered clips\")\n", "c = Counter([m[\"lang\"] for m in metas_map.values()])\n", "for k, v in c.most_common():\n", " print(k, c[k], round(sum(m[\"duration_s\"] for m in metas if m[\"lang\"] == k) / 60 / 60), \"hours\")" ] }, { "cell_type": "code", "execution_count": 11, "id": "c2ce5d00", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:06.963881Z", "start_time": "2023-11-05T18:46:06.960985Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "{'id': '3e73316d-6996-4afe-9b2a-3963b84c7472', 'lyrics': '[Intro, Producer tag: zolivagyok, 0xy Beat$, mun.grandi]\\n\\n[zolivagyok]\\n\\nGeci jó!\\n\\n[0xy Beat$]\\n\\n(söcinyitás, áháháhá geci)\\n\\n[mun.grandi]\\nA nődet kinyaltam, a pakkod elloptam\\nHa baszom a rezet, akkor vágom, hogy\\u2005jól\\u2005van\\nAz élet ér\\u2005véget, de tüstént, de nyomban\\nA laszti\\u2005most víz nélkül tombol a gyomromban\\n\\n[Verse 1: mun.grandi]\\n\\nElszakad a gumi, jön az AIDS-es baba\\nAz anyád kiállt lentről: “Könyörgöm, csak most ne hajba!”\\nGecis a kenyered, mert elmentem a vajba\\nMost éhes maradsz, geci, dögöljél meg most meg rajta!\\n\\nKellene már egy jó nagyot baszni\\nHogy holnap ne legyen full off, és fasz ki\\nA bráner már kék, nem kelt ki egy hete\\nMár robbannék szét, már reped szét a here!\\n\\n[Verse 2: zolivagyok]\\n\\nNyuggerek az esetem, festetthajú Fideszes\\nRusnya ráncos picsájukba boldogan beleverem\\nAz Erzsivel pultolok, a szájukba hugyozok\\nZoli vagyok, nem egy veréb, pinátokba hamuzok!\\n\\nZoli vagyok, a Hidegkút réme\\nHüviben ha meglátlak, életed vége\\nTekert Camelt szívok, buzi fake alterek\\nÉn trendsetter vagyok, ez a versem vége\\n\\n\\n[Hook: zolivagyok]\\n\\nJee, buzi fake alterek\\nJee! (Geci jó!)\\n\\n[Verse 3: $hroomlex The Lil Fuckup]\\n\\nA nődet annyit baszom, nem műkszik már a faszom\\nCrack houseban lakom, jön le már a plafon\\nTudják, hogy ott lakom, tűnyomos a karom\\nVérben forog a szemem, mikor anyádat baszom\\n\\nNem enyém a ház, de szinte benne lakom\\nAz anyád már képről ismeri a faszom\\nÍgy eszem a pinát, mint más a sajtos tésztát\\nMég egy pici crack, kettéfejelek egy téglát\\n\\nFegyvernek székláb, ételnek céklát\\nCsak látni jöttem anyád, nem értem, miért állt\\nMa megbasztam négy lányt, öreg anyád baszom\\nAzt mondja, valami fényt lát, de a farkam bele rakom... ez nem jön ki...\\n\\n[Hook 2: zolivagyok]\\n\\nJee, buzi fake alterek\\nA Hidegkút réme\\nÉn trendsetter vagyok\\nBuzi fake alterek! Jee! (Geci jó!)', 'duration_s': 156, 'lang': 'hu', 'audio_filepath': 's3://suno-data/datasets/harvest/genius_hq/audio/JKVVtomsOKg.webm', 'original_id': 'JKVVtomsOKg', 'genius_slug': '0xy-beat-edesharmas-lyrics', 'genius_views': 92}\n" ] } ], "source": [ "for m in metas:\n", " if m[\"lang\"] == \"hu\":\n", " print(m)\n", " break" ] }, { "cell_type": "code", "execution_count": 12, "id": "d7227e54", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:07.011901Z", "start_time": "2023-11-05T18:46:06.966784Z" } }, "outputs": [], "source": [ "# selected_metas = random.sample([m for m in metas if m[\"lang\"] == \"zh\"], k=1000)\n", "# print(len(selected_metas), len(set(m['original_id'] for m in selected_metas)))" ] }, { "cell_type": "code", "execution_count": 13, "id": "433b9ac9", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:07.059662Z", "start_time": "2023-11-05T18:46:07.013869Z" } }, "outputs": [], "source": [ "# with open(\"/home/tony/Data/Hoot/metas.json\", \"w\") as fp:\n", "# json.dump(metas, fp)" ] }, { "cell_type": "code", "execution_count": 14, "id": "6870cc25", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:07.106731Z", "start_time": "2023-11-05T18:46:07.062255Z" } }, "outputs": [], "source": [ "# {\"audio_filepath\": \"./an4/wav/an4_clstk/fash/an251-fash-b.wav\", \"duration\": 1.0, \"text\": \"yes\"}" ] }, { "cell_type": "code", "execution_count": 15, "id": "5cae0ba6", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:07.152049Z", "start_time": "2023-11-05T18:46:07.108647Z" } }, "outputs": [], "source": [ "# subset_meta = random.choices([m for m in metas if m[\"lang\"] == \"en\"], k=1000)" ] }, { "cell_type": "code", "execution_count": 16, "id": "71df5d56", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:07.469767Z", "start_time": "2023-11-05T18:46:07.153331Z" } }, "outputs": [], "source": [ "downloaded_files = os.listdir(\"/app/suno/data/hoot/\")" ] }, { "cell_type": "code", "execution_count": 17, "id": "be2444f7", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:07.810212Z", "start_time": "2023-11-05T18:46:07.471841Z" } }, "outputs": [], "source": [ "downloaded_ids = set([f.replace(\".wav\", \"\") for f in downloaded_files])" ] }, { "cell_type": "code", "execution_count": 18, "id": "c8227733", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:08.514968Z", "start_time": "2023-11-05T18:46:07.811787Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "984342\n" ] } ], "source": [ "subset_meta = [m for m in metas if (m[\"original_id\"] in downloaded_ids and m[\"lang\"] == \"en\")]\n", "print(len(subset_meta))\n", "# 986464" ] }, { "cell_type": "code", "execution_count": 19, "id": "2f9e4207", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:08.519641Z", "start_time": "2023-11-05T18:46:08.517003Z" } }, "outputs": [], "source": [ "def clean_text(text):\n", " text = text.replace(\"’\", \"'\").lower()\n", " text = text.replace(\"'\", \"\").replace('\"', \"\")\n", " text = re.sub(r\"\\[.+?\\]\", \" \", text)\n", " text = re.sub(r\"[\\d]\", \" \", text)\n", " text = re.sub(r\"▁\", \"\", text)\n", " # text = re.sub('[^a-zA-Z\\s]+', '', text)\n", " text = re.sub(r\"[^\\w]\", \" \", text)\n", " text = normalize_whitespace(text)\n", " return text\n", "# text = m[\"lyrics\"]\n", "# print(text, clean_text(text))" ] }, { "cell_type": "code", "execution_count": 20, "id": "1e61a956", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:08.567330Z", "start_time": "2023-11-05T18:46:08.521537Z" } }, "outputs": [], "source": [ "def _norm_lyrics_to_plain(text):\n", " text = text.lower()\n", " text = re.sub(r\"\\[.*?\\]\", \" \", text)\n", " text = re.sub(r\"[^a-z\\'\\s]\", \" \", text)\n", " return normalize_whitespace(text)" ] }, { "cell_type": "code", "execution_count": 21, "id": "6a5699a8", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:08.614537Z", "start_time": "2023-11-05T18:46:08.568629Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "directions in the exam angelyn 学校考试 准备好吗 老师正在发考试卷 老师讲解 这考试卷 讲完后就开始做考试卷 ferderick oh no惨了我没有为考试复习 我唯有去copy他人的考卷 是谁敢copy他人的考卷 老师就会拿起藤鞭 momo 哈 藤鞭来了 ferderick 等下你就吃藤鞭不用吃方便面 char kueh tiaw 加辣 boom boom momo 藤鞭来了了了了了了了了了 momo now 打我到打你 打我到打你 打我到打你 藤鞭来了 now 打我到打你 打我到打你 打我到打你 藤鞭来了 ferderick ya guys need to 复习for the 考试 拿着课本书包去学校诶 come on come on 复习 努力到死 need to 复习 不临时抱着佛脚 佛脚 去抄别人的考卷 考卷 一定要复习不管怎样都一定要考a考a angelyn 我惨了要拿考卷给爸爸看 可是怕他发现我没有考a 算了我唯有坦白地跟他讲 爸爸又会拿起藤鞭 momo 哈 藤鞭来了 ferderick 等下你就吃藤鞭不用吃方便面 炒干面加辣 boom boom momo 藤鞭来了了了了了了了了了 momo now 打我到打你 打我到打你 打我到打你 藤鞭来了 now 打我到打你 打我到打你 打我到打你 藤鞭来了 ferderick 我怕藤鞭所以我就要复习 angelyn 不管怎样我都一定要复习 ferderick 所以考试我就会考到了a all 不要吃藤鞭就要复习 yeah ferderick bring out the 复习本 all yeah directions 复复复复复习 复复复复复复习 momo 藤鞭来了 angelyn and ferderick 现在赶紧复习 考试就考到a 复复复复复习 复复复复复复习 现在赶紧复习 不然就会吃藤鞭\n", "direc tions in the exam angelyn ferderick oh no copy copy momo ferderick char kueh tiaw boom boom momo momo now now ferderick ya guys need to for the come on come on need to a a angelyn a momo ferderick boom boom momo momo now now ferderick angelyn ferderick a all yeah ferderick bring out the all yeah directions momo angelyn and ferderick a\n" ] } ], "source": [ "text = \"\"\"\n", "3 Direc▁tions In The ▁Exam\\n\\nAngelyn:\\n学校考试\\n准备好吗\\n老师正在发考试卷\\n老师讲解\\n这考试卷\\n讲完后就开始做考试卷\\nFerderick:\\nOh No惨了我没有为考试复习\\n我唯有去Copy他人的考卷\\n是谁敢Copy他人的考卷\\n老师就会拿起藤鞭\\nMomo:\\n哈！藤鞭来了\\nFerderick:\\n等下你就吃藤鞭不用吃方便面\\nChar Kueh Tiaw 加辣 Boom Boom\\nMomo:\\n藤鞭来了了了了了了了了了\\n\\nMomo:\\nNow\\n打我到打你\\n打我到打你\\n打我到打你\\n藤鞭来了\\nNow\\n打我到打你\\n打我到打你\\n打我到打你\\n藤鞭来了\\n\\nFerderick:\\nYa Guys Need To 复习For The 考试\\n拿着课本书包去学校诶\\nCome On Come On 复习\\n努力到死 Need To 复习\\n不临时抱着佛脚（佛脚）\\n去抄别人的考卷（考卷）\\n一定要复习不管怎样都一定要考A考A\\n\\n\\nAngelyn:\\n我惨了要拿考卷给爸爸看\\n可是怕他发现我没有考A\\n算了我唯有坦白地跟他讲\\n爸爸又会拿起藤鞭\\n\\nMomo:\\n哈！藤鞭来了\\nFerderick:\\n等下你就吃藤鞭不用吃方便面\\n炒干面加辣 Boom Boom\\nMomo:\\n藤鞭来了了了了了了了了了\\n\\nMomo:\\nNow\\n打我到打你\\n打我到打你\\n打我到打你\\n藤鞭来了\\nNow\\n打我到打你\\n打我到打你\\n打我到打你\\n藤鞭来了\\n\\n\\nFerderick:\\n我怕藤鞭所以我就要复习\\n\\nAngelyn:\\n不管怎样我都一定要复习\\n\\nFerderick:\\n所以考试我就会考到了A\\n\\nAll:\\n不要吃藤鞭就要复习\\nYeah\\n\\nFerderick:\\nBring Out The 复习本\\n\\nAll:\\nYeah\\n3 DIRECTIONS\\n复复复复复习\\n复复复复复复习\\n\\nMomo:\\n藤鞭来了\\n\\nAngelyn And Ferderick:\\n现在赶紧复习\\n考试就考到A\\n\\n复复复复复习\\n复复复复复复习\\n\\n现在赶紧复习\\n不然就会吃藤鞭\n", "\"\"\"\n", "print(clean_text(text))\n", "print(_norm_lyrics_to_plain(text))" ] }, { "cell_type": "code", "execution_count": 22, "id": "abc6bcee", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:46:09.063129Z", "start_time": "2023-11-05T18:46:08.615765Z" } }, "outputs": [ { "data": { "text/plain": [ "(984342, 984342)" ] }, "execution_count": 22, "metadata": {}, "output_type": "execute_result" } ], "source": [ "len(subset_meta), len(set(m[\"original_id\"] for m in subset_meta))" ] }, { "cell_type": "code", "execution_count": null, "id": "3a3246ca", "metadata": { "ExecuteTime": { "start_time": "2023-11-05T18:46:59.978Z" } }, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ " 47%|████████████████████████████████████████████████▏ | 460324/984342 [03:08<04:07, 2115.98it/s]" ] } ], "source": [ "info = []\n", "for meta in tqdm.tqdm(subset_meta):\n", " # output_path = os.path.join(\"/home/tony/Data/Hoot/exp\", f\"{meta['original_id']}.wav\")\n", " output_path = os.path.join(\"/app/suno/data/hoot\", f\"{meta['original_id']}.wav\")\n", " if meta[\"duration_s\"] > 240:\n", " continue\n", " # cleaned_text = clean_text(meta[\"lyrics\"])\n", " cleaned_text = _norm_lyrics_to_plain(meta[\"lyrics\"])\n", " # cleaned_tokens = tokenizer.encode(cleaned_text)\n", " # decoded_tokens = tokenizer.decode(cleaned_tokens)\n", " # if decoded_tokens != cleaned_text:\n", " # print(cleaned_text)\n", " # print(decoded_tokens)\n", " if len(cleaned_text) < 100:\n", " continue\n", " if len(cleaned_text) > meta[\"duration_s\"] * 12:\n", " # print(\"WTF\", cleaned_text)\n", " continue\n", " info.append(\n", " {\n", " \"audio_filepath\": output_path,\n", " \"duration\": meta[\"duration_s\"],\n", " \"text\": cleaned_text,\n", " }\n", " )\n", " if not os.path.exists(output_path):\n", " print(\"WTF\", output_path)\n", " continue\n", "# audio = Audio.from_s3(meta[\"audio_filepath\"])\n", "# audio = audio.convert(16_000, audio.byte_width, n_channels=1)\n", "# audio.to_wav(output_path)\n", "print(len(info))" ] }, { "cell_type": "code", "execution_count": null, "id": "618075a0", "metadata": { "ExecuteTime": { "start_time": "2023-11-05T18:47:02.544Z" } }, "outputs": [], "source": [ "print(\n", " sum(m[\"duration\"] for m in info) / 3600000,\n", " min(m[\"duration\"] for m in info),\n", " min(len(m[\"text\"]) for m in info),\n", " max(len(m[\"text\"]) for m in info),\n", ")" ] }, { "cell_type": "code", "execution_count": null, "id": "d8eb67a9", "metadata": { "ExecuteTime": { "start_time": "2023-11-05T18:47:03.112Z" } }, "outputs": [], "source": [ "cutoff = -1000\n", "random.shuffle(info)\n", "train_info = info[:cutoff]\n", "test_info = info[cutoff:]\n", "print(len(train_info), len(test_info))" ] }, { "cell_type": "code", "execution_count": null, "id": "a48b2d77", "metadata": { "ExecuteTime": { "end_time": "2023-11-03T15:08:42.921179Z", "start_time": "2023-11-03T15:08:42.921170Z" } }, "outputs": [], "source": [ "with open(\"/home/tony/Data/Hoot/en_train_manifest_fast_longer.json\", \"w\") as fp:\n", " # for l in train_info[:80]:\n", " for l in train_info[:80]:\n", " json.dump(l, fp, ensure_ascii=True)\n", " fp.write('\\n')\n", "with open(\"/home/tony/Data/Hoot/en_test_manifest_fast.json\", \"w\") as fp:\n", " for l in test_info[:8]:\n", " json.dump(l, fp, ensure_ascii=True)\n", " fp.write('\\n')" ] }, { "cell_type": "code", "execution_count": 89, "id": "4a7003ff", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T02:42:20.397177Z", "start_time": "2023-11-05T02:42:12.598297Z" } }, "outputs": [], "source": [ "with open(\"/home/tony/Data/Hoot/en_train_manifest_norm.json\", \"w\") as fp:\n", " for l in train_info:\n", " json.dump(l, fp, ensure_ascii=True)\n", " fp.write('\\n')\n", "with open(\"/home/tony/Data/Hoot/en_test_manifest_norm.json\", \"w\") as fp:\n", " for l in test_info:\n", " json.dump(l, fp, ensure_ascii=True)\n", " fp.write('\\n')" ] }, { "cell_type": "code", "execution_count": 25, "id": "506b6b20", "metadata": { "ExecuteTime": { "end_time": "2023-11-03T15:47:01.946046Z", "start_time": "2023-11-03T15:47:01.943030Z" } }, "outputs": [], "source": [ "x = []\n", "with open(\"/home/tony/Data/Hoot/en_train_manifest_fast_longer.json\", \"r\") as fp:\n", " for l in fp:\n", " x.append(json.loads(l))" ] }, { "cell_type": "code", "execution_count": null, "id": "5086e59d", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 44, "id": "253035d4", "metadata": { "ExecuteTime": { "end_time": "2023-11-03T19:04:38.398684Z", "start_time": "2023-11-03T19:04:38.396239Z" } }, "outputs": [], "source": [ "test_text = \"whoa oh oh whoa oh oh whoa oh oh oh oh oh oh whoa oh oh whoa oh oh whoa oh oh oh oh oh oh 취할 것 같은 불빛이 화려한 밤에 누구보다도 자유로워 지기를 바래 아름다운 이 순간이 끝나기 전에 바람소리에 이 음악이 실리길 원해 whoa oh oh whoa oh oh whoa oh oh oh oh oh oh whoa oh oh whoa oh oh whoa oh oh oh oh oh oh lets get it on tonight and set me free tonight 이 순간 즐겨봐 너와 나 소리 높여 lets groove it party time 두근거리는 느낌이 황홀한 밤에 푸른 바다에 달빛을 물들이길 바래 파도가 이 밤을 데려가기 전에 지금 이대로 시간이 멈추기를 원해 whoa oh oh whoa oh oh whoa oh oh oh oh oh oh whoa oh oh whoa oh oh whoa oh oh oh oh oh oh 이 순간 즐겨봐 너와 나 소리 높여 lets groove it party time whoa oh oh whoa oh oh whoa oh oh oh oh oh oh whoa oh oh whoa oh oh whoa oh oh oh oh oh oh 이 순간 즐겨봐 너와 나 소리 높여 lets groove it party time whoa whoa whoa yeah oh whoa oh whoa oh whoa\"" ] }, { "cell_type": "code", "execution_count": 45, "id": "9d5683b0", "metadata": { "ExecuteTime": { "end_time": "2023-11-03T19:04:38.634287Z", "start_time": "2023-11-03T19:04:38.632803Z" } }, "outputs": [], "source": [ "import sentencepiece" ] }, { "cell_type": "code", "execution_count": 49, "id": "65f394ae", "metadata": { "ExecuteTime": { "end_time": "2023-11-03T19:04:52.897296Z", "start_time": "2023-11-03T19:04:52.887433Z" } }, "outputs": [ { "data": { "text/plain": [ "True" ] }, "execution_count": 49, "metadata": {}, "output_type": "execute_result" } ], "source": [ "tokenizer = sentencepiece.SentencePieceProcessor()\n", "tokenizer.load(\"./tokenizers/en3/tokenizer_spe_bpe_v4096/tokenizer.model\")" ] }, { "cell_type": "code", "execution_count": 55, "id": "6baf7076", "metadata": { "ExecuteTime": { "end_time": "2023-11-03T19:32:07.782337Z", "start_time": "2023-11-03T19:32:07.778442Z" } }, "outputs": [ { "data": { "text/plain": [ "(329,)" ] }, "execution_count": 55, "metadata": {}, "output_type": "execute_result" } ], "source": [ "np.array(tokenizer.encode(test_text)).shape" ] }, { "cell_type": "code", "execution_count": 51, "id": "07cde444", "metadata": { "ExecuteTime": { "end_time": "2023-11-03T19:04:54.239964Z", "start_time": "2023-11-03T19:04:54.235510Z" } }, "outputs": [ { "data": { "text/plain": [ "'whoa oh oh whoa oh oh whoa oh oh oh oh oh oh whoa oh oh whoa oh oh whoa oh oh oh oh oh oh ⁇ 할 것 같은 불빛이 ⁇ 려한 밤에 누구보다도 자 ⁇ 로워 지기를 바래 아름다운 이 순간이 끝나기 전에 바람소리에 이 음 ⁇ 이 ⁇ 리길 원해 whoa oh oh whoa oh oh whoa oh oh oh oh oh oh whoa oh oh whoa oh oh whoa oh oh oh oh oh oh lets get it on tonight and set me free tonight 이 순간 ⁇ 겨봐 너와 나 소리 ⁇ 여 lets groove it party time 두 ⁇ 거리는 느 ⁇ 이 ⁇ 한 밤에 ⁇ 른 바다에 달빛을 물들이길 바래 ⁇ 도가 이 밤을 데려가기 전에 지금 이대로 시간이 ⁇ 기를 원해 whoa oh oh whoa oh oh whoa oh oh oh oh oh oh whoa oh oh whoa oh oh whoa oh oh oh oh oh oh 이 순간 ⁇ 겨봐 너와 나 소리 ⁇ 여 lets groove it party time whoa oh oh whoa oh oh whoa oh oh oh oh oh oh whoa oh oh whoa oh oh whoa oh oh oh oh oh oh 이 순간 ⁇ 겨봐 너와 나 소리 ⁇ 여 lets groove it party time whoa whoa whoa yeah oh whoa oh whoa oh whoa'" ] }, "execution_count": 51, "metadata": {}, "output_type": "execute_result" } ], "source": [ "tokenizer.decode(tokenizer.encode(test_text))" ] }, { "cell_type": "code", "execution_count": 52, "id": "2c5a702b", "metadata": { "ExecuteTime": { "end_time": "2023-11-03T19:09:29.365067Z", "start_time": "2023-11-03T19:09:29.362819Z" } }, "outputs": [], "source": [ "a = torch.ones([2, 100]).tolist()" ] }, { "cell_type": "code", "execution_count": 42, "id": "b50ec6b2", "metadata": { "ExecuteTime": { "end_time": "2023-11-03T16:11:46.392980Z", "start_time": "2023-11-03T16:11:46.390228Z" } }, "outputs": [ { "data": { "text/plain": [ "torch.Size([100])" ] }, "execution_count": 42, "metadata": {}, "output_type": "execute_result" } ], "source": [ "a = a.mean(axis=0)\n", "a.shape" ] }, { "cell_type": "code", "execution_count": 43, "id": "634bbbeb", "metadata": { "ExecuteTime": { "end_time": "2023-11-03T16:11:46.533290Z", "start_time": "2023-11-03T16:11:46.530540Z" } }, "outputs": [ { "data": { "text/plain": [ "torch.Size([100])" ] }, "execution_count": 43, "metadata": {}, "output_type": "execute_result" } ], "source": [ "a.reshape(-1).shape" ] }, { "cell_type": "code", "execution_count": null, "id": "eab66c15", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "792508c0", "metadata": { "ExecuteTime": { "end_time": "2023-11-03T15:08:42.925074Z", "start_time": "2023-11-03T15:08:42.925066Z" } }, "outputs": [], "source": [ "python ./scripts/process_asr_text_tokenizer.py \\\n", " --manifest=\"/home/tony/Data/Hoot/en_train_manifest.json\" \\\n", " --data_root=\"./tokenizers/en2/\" \\\n", " --vocab_size=1024 \\\n", " --tokenizer=\"spe\" \\\n", " --spe_character_coverage=0.998 \\\n", " --spe_type=\"bpe\" \\\n", " --log" ] }, { "cell_type": "code", "execution_count": 8, "id": "030401c8", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:54:19.781622Z", "start_time": "2023-11-05T18:54:19.778205Z" } }, "outputs": [ { "data": { "text/plain": [ "{'id': '38fbe48a-9c9a-4888-a15c-8dba4d33f037',\n", " 'lyrics': \"It's been a long time since I've met you\\nWe are somewhat close people\\nCome up to me and say hello\\nTell me your story\\n\\nWho did you meet and what did you do?\\nHow are you doing these days?\\nAfter exchanging news and updates\\nMy worries are over\\n\\nThen I'm glad I laughed\\nI'm fine, there's no problem\\nHowever, the truth is, sadly, I'm not fine\\nEvery night, every night, when I close my eyes I hate to die in your dreams\\n\\nI am unhappy to hear you hint about\\nA friend, who speaks badly to you\\nNo, now when I look back, now and again\\nBut rather, I'm thankful\\n\\nI'm glad that I'm laughing again\\nI'm fine, there's no problem\\n\\nHowever, the truth is, sadly, I'm not fine\\nEvery night, every night, I see you in my dreams without fail\\nHowever, everyone knows I'm not okay\\nEvery night when I close my eyes I hate to die in your dreams\\n\\nThen what can I do?\\nDay by day, the lies pile up\",\n", " 'duration_s': 231,\n", " 'lang': 'en',\n", " 'audio_filepath': 's3://suno-data/datasets/harvest/genius_hq/audio/uW03kXO1Cp0.webm',\n", " 'original_id': 'uW03kXO1Cp0',\n", " 'genius_slug': '10cm-10cm-however-english-translation-lyrics',\n", " 'genius_views': 35}" ] }, "execution_count": 8, "metadata": {}, "output_type": "execute_result" } ], "source": [ "metas[2]" ] }, { "cell_type": "code", "execution_count": 9, "id": "0b3b2981", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:54:25.710668Z", "start_time": "2023-11-05T18:54:25.677748Z" } }, "outputs": [], "source": [ "test_audio = Audio.from_file(\"/app/suno/data/hoot/uW03kXO1Cp0.wav\")" ] }, { "cell_type": "code", "execution_count": 10, "id": "8bc4fd0b", "metadata": { "ExecuteTime": { "end_time": "2023-11-05T18:54:31.641990Z", "start_time": "2023-11-05T18:54:30.861945Z" } }, "outputs": [ { "data": { "text/html": [ "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "test_audio.play()" ] }, { "cell_type": "code", "execution_count": null, "id": "9ffe94ae", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "353d4d0f", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "e3b77911", "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": "3.10.12" }, "toc": { "base_numbering": 1, "nav_menu": {}, "number_sections": true, "sideBar": true, "skip_h1_title": false, "title_cell": "Table of Contents", "title_sidebar": "Contents", "toc_cell": false, "toc_position": {}, "toc_section_display": true, "toc_window_display": false } }, "nbformat": 4, "nbformat_minor": 5 }