{ "cells": [ { "cell_type": "markdown", "id": "de426cde", "metadata": {}, "source": [ "### Fetch episodes" ] }, { "cell_type": "code", "execution_count": null, "id": "a54ce4b3", "metadata": {}, "outputs": [], "source": [ "python /home/georg/code/glockenspiel/suno_utils/suno_utils/scripts/fetch_podcasts.py \\\n", " --podcast_db_filepath=\"/mnt/data-ssd-1/data/podcasts/meta/podcastindex_feeds.db\" \\\n", " --podcast_ids=\"2782546,931201,1665134,3523330,3177006,243005,5005075,829896,488174,4284986\" \\\n", " --output_dir=\"/mnt/data-ssd-1/data/podcasts/audio\" \\\n", " --last_n_episodes=1" ] }, { "cell_type": "code", "execution_count": null, "id": "1e1f1510", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "d0c8ddcc", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "8a4af0ef", "metadata": {}, "source": [ "### 00 - Make rev manifest" ] }, { "cell_type": "code", "execution_count": 1, "id": "4b89af38", "metadata": {}, "outputs": [], "source": [ "from suno_utils.utils.podcasts import get_episodes_to_label\n", "import numpy as np\n", "\n", "PODCAST_IDS = set([\n", " \"2782546\", \"931201\", \"1665134\", \"3523330\", \"3177006\", \n", " \"243005\", \"829896\", \"488174\", \"4284986\"#, \"5005075\"\n", "])\n", "\n", "PODCAST_AUDIO_DIR = \"/mnt/data-ssd-1/data/podcasts/audio/\"" ] }, { "cell_type": "code", "execution_count": 4, "id": "a6a1e365", "metadata": {}, "outputs": [], "source": [ "rev_manifest = []\n", "for podcast_id in PODCAST_IDS:\n", " audio_dir = PODCAST_AUDIO_DIR + f\"{podcast_id}/\"\n", " annotation_manifest = get_episodes_to_label(\n", " audio_dir, \n", " min_episode_duration=5*60, \n", " min_total_duration_s=5*60, \n", " max_per_episode_duration_s=5*60, \n", " shuffle=False,\n", " )\n", " rev_manifest.extend(annotation_manifest)\n", "assert(np.sum([e[2] - e[1] for e in rev_manifest]) == 5 * 60 * len(podcast_ids))" ] }, { "cell_type": "code", "execution_count": 7, "id": "a390ebfe", "metadata": {}, "outputs": [], "source": [ "import uuid\n", "from suno_utils.utils.text import write_jsonl\n", "\n", "manifest_data = [\n", " {\"filepath\": e[0], \"start_s\": e[1], \"end_s\": e[2], \"hotwords\": e[3], \"uuid\": str(uuid.uuid4())} \n", " for e in rev_manifest\n", "]\n", "write_jsonl(manifest_data, \"fili_100/pre_pipeline/fili_100_rev_manifest.jsonl\")" ] }, { "cell_type": "code", "execution_count": 8, "id": "acf8da05", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "{\"filepath\": \"/mnt/data-ssd-1/data/podcasts/audio/829896/0d73ac8e-40de-49b8-9eb5-282fc7edb81d.m4a\", \"start_s\": 0, \"end_s\": 300.0, \"hotwords\": [\"Dahon\", \"Filipino\", \"Leaf\", \"English\"], \"uuid\": \"fa0564cf-7847-4dfb-a8bb-35d881eb0b06\"}\r\n", "{\"filepath\": \"/mnt/data-ssd-1/data/podcasts/audio/243005/fa46a049-4749-9993-b7b5-90035ceb40af.mp3\", \"start_s\": 0, \"end_s\": 300.0, \"hotwords\": [\"Philippines\", \"Spotify\", \"Jason Shaw\"], \"uuid\": \"bf6d9a07-253c-4400-8157-ad556b74579f\"}\r\n" ] } ], "source": [ "!head -2 fili_100/pre_pipeline/fili_100_rev_manifest.jsonl" ] }, { "cell_type": "code", "execution_count": null, "id": "d598b979", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "3cfb42b7", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "1e31e896", "metadata": {}, "source": [ "### (optional) listen to samples" ] }, { "cell_type": "code", "execution_count": 11, "id": "a3f8c8af", "metadata": {}, "outputs": [], "source": [ "import random\n", "from suno_utils.utils.text import load_jsonl\n", "from suno_utils.utils.notebook import Audio\n", "\n", "manifest_data = load_jsonl(\"fili_100/pre_pipeline/fili_100_rev_manifest.jsonl\")\n", "test_manifest = manifest_data[:]" ] }, { "cell_type": "code", "execution_count": null, "id": "52a91856", "metadata": {}, "outputs": [], "source": [ "random.shuffle(test_manifest)\n", "for e in test_manifest[:5]:\n", " print(\"-\"*10)\n", " print(e[\"filepath\"])\n", " start_s = round(e[\"start_s\"] + random.random() * (e[\"end_s\"] - e[\"start_s\"]), 1)\n", " end_s = round(np.min([start_s + 30, e[\"end_s\"]]), 1)\n", " print(start_s, end_s)\n", " Audio.from_file(e[\"filepath\"]).play(start_s, end_s)" ] }, { "cell_type": "code", "execution_count": null, "id": "353541e9", "metadata": {}, "outputs": [], "source": [ "# us accent in:\n", "# /mnt/data-ssd-1/data/podcasts/audio/488174/dde8118f-ef11-484a-a957-52b10a05399b.mp3\n", "# /mnt/data-ssd-1/data/podcasts/audio/931201/19a4e362-b732-4525-a5a8-95ad58ff94c0.mp3\n", "\n", "# some filipino\n", "# /mnt/data-ssd-1/data/podcasts/audio/3523330/49bdaac0-92d9-4ba5-831b-b6663fefea4d.m4a" ] }, { "cell_type": "code", "execution_count": null, "id": "e4cd4618", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "b16a0dd0", "metadata": {}, "source": [ "### 01 - upload items" ] }, { "cell_type": "code", "execution_count": 28, "id": "cb256e57", "metadata": {}, "outputs": [], "source": [ "import tqdm\n", "from suno_utils.utils.text import load_jsonl\n", "from suno_utils.audio import Audio\n", "\n", "FILI_100_DIR = \"/home/georg/notebooks/datasets/phillippines/fili_100/pre_pipeline/\"\n", "manifest_data = load_jsonl(FILI_100_DIR + \"00_rev_manifest.jsonl\")" ] }, { "cell_type": "code", "execution_count": null, "id": "2453089e", "metadata": {}, "outputs": [], "source": [ "import json\n", "from suno_utils.web.rev import get_auth_string, upload_file\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": 3, "id": "8c66f7f9", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "100%|██████████| 9/9 [01:12<00:00, 8.02s/it]\n" ] } ], "source": [ "order_items = []\n", "for audio_meta in tqdm.tqdm(manifest_data):\n", " audio = Audio.from_file(audio_meta[\"filepath\"])\n", " audio = audio.get_slice(audio_meta[\"start_s\"], audio_meta[\"end_s\"])\n", " audio = audio.convert(44_100, 2, 1)\n", " wav_audio_filepath = f\"{FILI_100_DIR}rev_files/{audio_meta['uuid']}.wav\"\n", " audio.to_wav(wav_audio_filepath)\n", " mp3_audio_filepath = f\"{FILI_100_DIR}rev_files/{audio_meta['uuid']}.mp3\"\n", " audio.to_mp3(mp3_audio_filepath)\n", " media_loc = upload_file(AUTH_STR, audio_filepath, file_ref_str=audio_meta[\"uuid\"])\n", " order_items.append({\n", " \"media_loc\": media_loc,\n", " \"hotwords\": audio_meta[\"hotwords\"],\n", "# \"accents\": [\"Australian\"],\n", " \"duration_s\": int(round(audio.duration_s)),\n", " \"uuid\": audio_meta[\"uuid\"],\n", " })" ] }, { "cell_type": "code", "execution_count": 12, "id": "a0ee5f4f", "metadata": {}, "outputs": [], "source": [ "with open(FILI_100_DIR + \"01_order_items.json\", \"w\") as f:\n", " json.dump(order_items, f)" ] }, { "cell_type": "code", "execution_count": null, "id": "219e9212", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "ba640695", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "9131358e", "metadata": {}, "source": [ "### 02 - make order" ] }, { "cell_type": "code", "execution_count": null, "id": "7ccf6e0a", "metadata": {}, "outputs": [], "source": [ "import json\n", "from suno_utils.web.rev import get_auth_string, make_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": 3, "id": "0b300ef5", "metadata": {}, "outputs": [], "source": [ "FILI_100_DIR = \"/home/georg/notebooks/datasets/phillippines/fili_100/pre_pipeline/\"\n", "\n", "with open(FILI_100_DIR + \"01_order_items.json\", \"r\") as f:\n", " order_items = json.load(f)" ] }, { "cell_type": "code", "execution_count": 4, "id": "8318e46f", "metadata": {}, "outputs": [], "source": [ "order_number = make_order(AUTH_STR, order_items, order_ref_str=\"fili_100\")" ] }, { "cell_type": "code", "execution_count": 6, "id": "27549d51", "metadata": {}, "outputs": [], "source": [ "with open(FILI_100_DIR + \"02_order_numer.txt\", \"w\") as f:\n", " f.write(order_number)" ] }, { "cell_type": "code", "execution_count": null, "id": "210eae71", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "f30acc60", "metadata": {}, "source": [ "### 04 - pull transcripts" ] }, { "cell_type": "code", "execution_count": 1, "id": "65101ca4", "metadata": {}, "outputs": [], "source": [ "import json\n", "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": 2, "id": "de613d78", "metadata": {}, "outputs": [], "source": [ "import os\n", "\n", "FILI_100_DIR = \"/home/georg/notebooks/datasets/phillippines/fili_100/pre_pipeline/\"\n", "\n", "with open(FILI_100_DIR + \"02_order_numer.txt\") as f:\n", " order_number = f.read()" ] }, { "cell_type": "code", "execution_count": null, "id": "498cc9fc", "metadata": {}, "outputs": [], "source": [ "transcript_metas = get_finished_order(AUTH_STR, order_number, blocking=True)" ] }, { "cell_type": "code", "execution_count": 21, "id": "0c49ce69", "metadata": {}, "outputs": [], "source": [ "for transcript_meta in transcript_metas:\n", " transcript_json = get_transcript(AUTH_STR, transcript_meta[\"rev_id\"])\n", " transcript_filepath = os.path.join(FILI_100_DIR, \"rev_files\", f\"{transcript_meta['uuid']}.json\")\n", " with open(transcript_filepath, \"w\") as f:\n", " json.dump(transcript_json, f)" ] }, { "cell_type": "code", "execution_count": 45, "id": "3cd62557", "metadata": {}, "outputs": [], "source": [ "annotation_manifest = [\n", " {\n", " \"audio_filepath\": os.path.join(FILI_100_DIR, \"rev_files\", f\"{transcript_meta['uuid']}.wav\"),\n", " \"transcript_filepath\": os.path.join(FILI_100_DIR, \"rev_files\", f\"{transcript_meta['uuid']}.json\"),\n", " }\n", " for transcript_meta in transcript_metas\n", "]\n", "assert(all([\n", " os.path.exists(e[\"audio_filepath\"]) and os.path.exists(e[\"transcript_filepath\"]) \n", " for e in annotation_manifest\n", "]))" ] }, { "cell_type": "code", "execution_count": 46, "id": "bce65300", "metadata": {}, "outputs": [], "source": [ "with open(FILI_100_DIR + \"03_annotation_manifest.json\", \"w\") as f:\n", " json.dump(annotation_manifest, f)" ] }, { "cell_type": "code", "execution_count": null, "id": "1f3a5a1d", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "11958a17", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "82680cf5", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "a05589e2", "metadata": {}, "source": [ "### create dataset" ] }, { "cell_type": "code", "execution_count": null, "id": "4ebd702c", "metadata": {}, "outputs": [], "source": [ "# run the pipeline\n", "python -m suno_utils.scripts.rev_podcast.run \\\n", " --output_dir=\"/home/georg/notebooks/datasets/phillippines/fili_100/pipeline\" \\\n", " --use_gpu=True \\\n", " --conda-env=\"mfa\" \\\n", " --conda_sh_filepath=\"/home/georg/anaconda3/etc/profile.d/conda.sh\" \\\n", " --mfa_dictionary_name=\"english\" \\\n", " --mfa_acoustic_model_name=\"english\" \\\n", " --num_cores_mfa=20 \\\n", " --manifest_file=\"/home/georg/notebooks/datasets/phillippines/fili_100/pre_pipeline/03_annotation_manifest.json\" \\\n", " --dataset_name=\"fili_100\" \\\n", " --sample_rate=44100 \\\n", " --s3_sample_data_prefix=\"s3://suno-static-public/datasets/fili-100/sample-data/segments/\"" ] }, { "cell_type": "code", "execution_count": 51, "id": "efbed978", "metadata": {}, "outputs": [], "source": [ "# # remove sample 5\n", "# import json\n", "# with open(\"/tmp/georg_southern/artifacts/web-sample-data/sample_manifest_web.json\") as f:\n", "# d = json.load(f)\n", "# del d[\"sample\"][5]\n", "# with open(\"/tmp/georg_southern/artifacts/web-sample-data/sample_manifest_web.json\", \"w\") as f:\n", "# json.dump(d, f)" ] }, { "cell_type": "code", "execution_count": null, "id": "49100414", "metadata": {}, "outputs": [], "source": [ "# # move to a sane place if it worked\n", "# mv /tmp/georg_southern /mnt/data-ssd-1/data/southern-comfort/output\n", "\n", "# # upload the downloadable samples\n", "# aws s3 cp \\\n", "# --acl public-read /mnt/data-ssd-1/data/southern-comfort/output/artifacts/data_sample.tar.gz \\\n", "# s3://suno-static-public/datasets/southern-comfort/southern_comfort_sample.tar.gz\n", "\n", "# # upload the web FE samples\n", "# aws s3 cp \\\n", "# --acl public-read \\\n", "# --recursive /mnt/data-ssd-1/data/southern-comfort/output/artifacts/web-sample-data/ \\\n", "# s3://suno-static-public/datasets/southern-comfort/sample-data" ] }, { "cell_type": "code", "execution_count": null, "id": "62f22204", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "1eba47b2", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "5e375b14", "metadata": {}, "source": [ "## Make sanas data" ] }, { "cell_type": "code", "execution_count": 54, "id": "e9084aa9", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "71 segments remaining\n" ] } ], "source": [ "import json\n", "import random\n", "import uuid\n", "import re\n", "from suno_utils.utils.text import load_jsonl\n", "from suno_utils.audio import Audio, Tokens\n", "\n", "FILI_100_DIR = \"/home/georg/notebooks/datasets/phillippines/fili_100/pipeline/\"\n", "# data = load_jsonl(FILI_100_DIR + \"artifacts/04_segment_meta_distribution.jsonl\")\n", "with open(FILI_100_DIR + \"artifacts/03_segment_meta_post_timestamp.json\") as f:\n", " data = json.load(f)\n", "random.seed(6006)\n", "random.shuffle(data)\n", "\n", "keys_to_scrub = (\n", " \"source_audio_uri\",\n", " \"source_audio_id\",\n", " \"source_transcript_id\",\n", " \"source_transcript_uri\",\n", " \"transcript_asr\",\n", " \"source_audio_offset_s\",\n", " \"ignore_segment\",\n", " \"ignore_reason\",\n", ")\n", "\n", "sub_data = []\n", "for e in data:\n", " tokens = Tokens.from_dict(e[\"transcript\"][\"tokens\"])\n", " if tokens.n_speakers > 1:\n", " continue\n", " if e[\"ignore_segment\"]:\n", " continue\n", " e[\"duration_s\"] = round(e[\"duration_s\"], 2)\n", " for ee in e[\"transcript\"][\"tokens\"]:\n", " if \"speaker_id\" in ee:\n", " ee[\"speaker_id\"] = str(uuid.uuid3(uuid.UUID(e[\"source_audio_id\"]), str(ee[\"speaker_id\"])))\n", " e[\"uri\"] = re.search(r\"segments\\/.+$\", e[\"uri\"]).group(0)\n", " for key in keys_to_scrub:\n", " if key in e:\n", " del e[key]\n", " sub_data.append(e)\n", "print(len(sub_data), \"segments remaining\")" ] }, { "cell_type": "code", "execution_count": 59, "id": "ac98cec3", "metadata": {}, "outputs": [], "source": [ "with open(FILI_100_DIR + \"artifacts/fili_sample_manifest.json\", \"w\") as f:\n", " json.dump(sub_data, f)" ] }, { "cell_type": "code", "execution_count": null, "id": "2ecbed1e", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "6f1ba8b2", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "b3c64c20", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "fbd28c43", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "47baa98d", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "8ea135d1", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "30e6013b", "metadata": {}, "source": [ "## (optional) listen to examples" ] }, { "cell_type": "code", "execution_count": null, "id": "672d4159", "metadata": {}, "outputs": [], "source": [ "import random\n", "from suno_utils.utils.text import load_jsonl\n", "from suno_utils.utils.notebook import Audio\n", "\n", "FILI_100_DIR = \"/home/georg/notebooks/datasets/phillippines/fili_100/pipeline/\"\n", "# data = load_jsonl(FILI_100_DIR + \"artifacts/04_segment_meta_distribution.jsonl\")\n", "with open(FILI_100_DIR + \"artifacts/03_segment_meta_post_timestamp.json\") as f:\n", " data = json.load(f)\n", "random.seed(6006)\n", "random.shuffle(data)" ] }, { "cell_type": "code", "execution_count": 12, "id": "3435319c", "metadata": {}, "outputs": [], "source": [ "# eca23ce6-b49b-46d7-9980-2c7cec3c920e: edge mistake" ] }, { "cell_type": "code", "execution_count": null, "id": "6231d935", "metadata": {}, "outputs": [], "source": [ "for e in data[:10]:\n", " print(e[\"id\"])\n", " print(\" \" + e[\"transcript_asr\"][\"transcript\"])\n", " print(\" \" + e[\"transcript_normalized\"][\"text\"])\n", " print(\" \" + e[\"transcript\"][\"text\"])\n", " Audio.from_file(e[\"uri\"]).play()\n", " print(\"-\"*10)" ] }, { "cell_type": "code", "execution_count": 20, "id": "b14b7c68", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "100%|██████████| 136/136 [00:00<00:00, 9723.77it/s]\n" ] } ], "source": [ "import tqdm\n", "import pandas as pd\n", "from suno_utils.utils.metrics import get_wer, get_cer\n", "\n", "m_data = []\n", "for e in tqdm.tqdm(data):\n", " wer = get_wer(e[\"transcript_normalized\"][\"text\"], e[\"transcript_asr\"][\"transcript\"])\n", " cer = get_cer(e[\"transcript_normalized\"][\"text\"], e[\"transcript_asr\"][\"transcript\"])\n", " m_data.append((\n", " e[\"id\"],\n", " e[\"transcript_asr\"][\"transcript\"],\n", " e[\"transcript_normalized\"][\"text\"],\n", " e[\"transcript_asr\"][\"transcript\"],\n", " e[\"uri\"],\n", " wer,\n", " cer,\n", " ))\n", "metrics_df = pd.DataFrame(m_data, columns=[\"id\", \"pred\", \"true\", \"raw_true\", \"uri\", \"wer\", \"cer\"])\n", "metrics_df = metrics_df.sort_values(by=\"cer\", ascending=False).reset_index(drop=True)" ] }, { "cell_type": "code", "execution_count": null, "id": "8a269ed9", "metadata": {}, "outputs": [], "source": [ "for _, row in metrics_df.head(10).tail(5).iterrows():\n", " print(row[\"id\"], \"-\", round(row[\"cer\"], 3), \"-\", round(row[\"wer\"], 3))\n", " print(\" \" + row[\"pred\"])\n", " print(\" \" + row[\"true\"])\n", " Audio.from_file(row[\"uri\"]).play()\n", " print(\"-\"*10)" ] }, { "cell_type": "code", "execution_count": null, "id": "6dec1951", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "63255f60", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "ee77181a", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "dce973c4", "metadata": {}, "source": [ "### Playground" ] }, { "cell_type": "code", "execution_count": 2, "id": "9473b0e7", "metadata": {}, "outputs": [], "source": [ "import json" ] }, { "cell_type": "code", "execution_count": 51, "id": "c15e33c1", "metadata": {}, "outputs": [], "source": [ "# make mini manifest\n", "with open(\"/home/georg/notebooks/datasets/phillippines/fili_100/pre_pipeline/03_annotation_manifest.json\") as f:\n", " m = json.load(f)\n", "with open(\"/tmp/manifest.json\", \"w\") as f:\n", "# json.dump(m[8:9], f)\n", " json.dump(m[8:9], f)" ] }, { "cell_type": "code", "execution_count": 18, "id": "9add6138", "metadata": {}, "outputs": [], "source": [ "# m[8:9]: WARNING:suno_utils.datasets.rev_podcast.process_data:only 34.3% segment coverage for /home/georg/notebooks/datasets/phillippines/fili_100/pre_pipeline/rev_files/c7a5d271-5ff9-470f-85af-5e9c9e2f8465.json" ] }, { "cell_type": "code", "execution_count": null, "id": "809ec352", "metadata": {}, "outputs": [], "source": [ "python -m suno_utils.scripts.rev_podcast.run \\\n", " --output_dir=\"/tmp/blabla\" \\\n", " --use_gpu=True \\\n", " --conda-env=\"mfa\" \\\n", " --conda_sh_filepath=\"/home/georg/anaconda3/etc/profile.d/conda.sh\" \\\n", " --mfa_dictionary_name=\"english\" \\\n", " --mfa_acoustic_model_name=\"english\" \\\n", " --num_cores_mfa=20 \\\n", " --manifest_file=\"/tmp/manifest.json\" \\\n", " --dataset_name=\"fili_100\" \\\n", " --sample_rate=44100 \\\n", " --s3_sample_data_prefix=\"s3://suno-static-public/datasets/fili-100/sample-data/segments/\"" ] }, { "cell_type": "code", "execution_count": 70, "id": "f2207c7f", "metadata": {}, "outputs": [], "source": [ "# with open(\"/tmp/blabla/artifacts/02_segment_meta_post_norm_and_asr.json\") as f:\n", "# d2 = json.load(f)\n", "# with open(\"/tmp/blabla/artifacts/03_segment_meta_post_timestamp.json\") as f:\n", "# d3 = json.load(f)" ] }, { "cell_type": "code", "execution_count": 37, "id": "28b1ebe2", "metadata": {}, "outputs": [], "source": [ "# for n, e in enumerate(d3):\n", "# if e[\"ignore_segment\"]:\n", "# print(n)" ] }, { "cell_type": "code", "execution_count": 1, "id": "8bb82ee9", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "[NeMo W 2022-06-05 16:36:58 experimental:27] Module is experimental, not ready for production and is not fully supported. Use at your own risk.\n", "[NeMo W 2022-06-05 16:36:58 experimental:27] Module is experimental, not ready for production and is not fully supported. Use at your own risk.\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "[NeMo I 2022-06-05 16:36:59 post_processing:56] Post processing graph was restored from /tmp/prod_nemo_fst/en_tn_post_processing.far.\n", "[NeMo I 2022-06-05 16:37:09 tokenize_and_classify_with_audio:105] ClassifyFst.fst was restored from /tmp/prod_nemo_fst/_cased_en_tn_False_deterministicnemo_custom_norm.tsv.far.\n", "[NeMo I 2022-06-05 16:37:09 verbalize_final:58] VerbalizeFinalFst graph was restored from /tmp/prod_nemo_fst/en_tn_False_deterministic_verbalizer.far.\n" ] } ], "source": [ "import json\n", "from suno_utils.utils.notebook import Audio\n", "from suno_utils.datasets.rev_podcast.process_data import (\n", " _get_podcast_segments_gentle, parse_transcript, postprocess_align_dict\n", ")\n", "from suno_utils.utils.text import load_json\n", "\n", "uuid = \"c7a5d271-5ff9-470f-85af-5e9c9e2f8465\"\n", "\n", "d = \"/home/georg/notebooks/datasets/phillippines/fili_100/pre_pipeline/rev_files/\"\n", "\n", "audio_filepath = d + f\"{uuid}.wav\"\n", "transcript_filepath = d + f\"{uuid}.json\"\n", "gentle_filepath = f\"/tmp/blabla/alignment-gentle/{uuid}.json\"" ] }, { "cell_type": "code", "execution_count": 2, "id": "a6a7d02a", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "2 segments\n", "7.1 % coverage\n" ] } ], "source": [ "transcript = load_json(transcript_filepath)\n", "speaker_map, raw_tokens = parse_transcript(transcript)\n", "with open(gentle_filepath) as fi:\n", " align_dict = json.load(fi)\n", "aligned_tokens = postprocess_align_dict(align_dict, tokens_with_tags=raw_tokens)\n", "audio = Audio.from_file(audio_filepath)\n", "segments = _get_podcast_segments_gentle(\n", " audio, transcript, gentle_filepath,\n", " remove_background_for_align=True,\n", " remove_background_for_vad=True,\n", ")\n", "print(len(segments), \"segments\")\n", "print(round(sum([e[1][1] - e[1][0] for e in segments]) / audio.duration_s * 100, 1), \"% coverage\")\n", "# 9 segments\n", "# 21.5 % coverage" ] }, { "cell_type": "code", "execution_count": 3, "id": "bcfca340", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "(238.48, 241.23)\n", "I recommend you write it down\n" ] }, { "data": { "text/html": [ "\n", " \n", " " ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "----------\n", "(281.36, 300.0)\n", "ah, it's Elder Scrolls and, ah, the oldest show series mostly, but I do already stream -- I work every day Mondays through Fridays [laughter] See\n" ] }, { "data": { "text/html": [ "\n", " \n", " " ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "----------\n" ] } ], "source": [ "for s in segments[:5]:\n", " print(s[1])\n", " print(s[-1].text)\n", " s[0].play()\n", " print(\"-\"*10)" ] }, { "cell_type": "code", "execution_count": null, "id": "db271a5e", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "f6dac7cd", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 2, "id": "7745aecc", "metadata": {}, "outputs": [], "source": [ "!cp fili_100/pipeline/artifacts/data_sample.tar.gz ." ] }, { "cell_type": "code", "execution_count": 7, "id": "426723e2", "metadata": {}, "outputs": [], "source": [ "!tar -xf data_sample.tar.gz" ] }, { "cell_type": "code", "execution_count": 11, "id": "74ad08cd", "metadata": {}, "outputs": [], "source": [ "!mkdir fili_sample_data" ] }, { "cell_type": "code", "execution_count": 13, "id": "e9909cc9", "metadata": {}, "outputs": [], "source": [ "!tar -czf fili_sample_data.tar.gz fili_sample_data" ] }, { "cell_type": "code", "execution_count": null, "id": "85016617", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "98a16801", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "679fb57f", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "862b93ca", "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 }