{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "45c1480d",
"metadata": {},
"outputs": [],
"source": [
"import numpy as np\n",
"import pandas as pd\n",
"\n",
"%matplotlib inline\n",
"\n",
"import os\n",
"import json\n",
"import requests\n",
"import urllib3\n",
"\n",
"from bs4 import BeautifulSoup, SoupStrainer\n",
"import bs4\n",
"import time\n",
"\n",
"import datetime\n",
"import tqdm\n",
"import glob\n",
"\n",
"\n",
"from suno_utils.utils.conversion import load_audio_as_array, write_wav, convert_audio\n",
"from suno_utils.utils.notebook import play_audio, play_array\n",
"from IPython.display import Audio\n",
"\n",
"from suno_utils.utils.conversion import load_audio_as_bytes\n",
"from suno_utils.utils.alignment import (\n",
" align_segments,\n",
" process_align_meta,\n",
" postprocess_slice_meta,\n",
")\n",
"\n",
"from suno_utils.utils.slicer import (\n",
" get_align_gaps,\n",
" get_vad_gaps,\n",
" get_intervals_from_gaps,\n",
" get_audio_slice,\n",
")\n",
"from suno_utils.utils.alignment import _reconstitute_slice_text\n",
"import portion\n",
"import time\n",
"from suno_utils.utils.alignment import postprocess_slice_meta"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f026efe4",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 170,
"id": "6ad66e99",
"metadata": {},
"outputs": [],
"source": [
"BAD_DATE_STRS = {\n",
" \"20220216\", # corrupt audio \n",
" \"20211118\" # missing files\n",
"} \n",
"base_dir = \"/mnt/data-ssd-1/data/nz-parl/\"\n",
"manifest_file = os.path.join(base_dir, \"manifest.json\")"
]
},
{
"cell_type": "code",
"execution_count": 192,
"id": "29f932ec",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"25\n"
]
}
],
"source": [
"with open(manifest_file, \"r\") as fi:\n",
" manifest_tmp = json.load(fi)[:32]#[:23]\n",
"\n",
"manifest = []\n",
"seen = set()\n",
"for m in manifest_tmp:\n",
" ds = m[\"url\"].split(\"_\")[-1]\n",
" if ds in seen or ds in BAD_DATE_STRS:\n",
" continue\n",
" seen.add(ds)\n",
" manifest.append(m)\n",
"print(len(manifest))\n",
"del manifest_tmp"
]
},
{
"cell_type": "code",
"execution_count": 193,
"id": "79b7dc56",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"['20220317',\n",
" '20220316',\n",
" '20220315',\n",
" '20220310',\n",
" '20220309',\n",
" '20220308',\n",
" '20220303',\n",
" '20220302',\n",
" '20220301',\n",
" '20220217',\n",
" '20220215',\n",
" '20220210',\n",
" '20220209',\n",
" '20220208',\n",
" '20211215',\n",
" '20211214',\n",
" '20211209',\n",
" '20211208',\n",
" '20211207',\n",
" '20211125',\n",
" '20211123',\n",
" '20211117',\n",
" '20211116',\n",
" '20211111',\n",
" '20211110']"
]
},
"execution_count": 193,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"[m[\"url\"].split(\"_\")[-1] for m in manifest]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ded0ef14",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"id": "0308dcbc",
"metadata": {},
"source": [
"## Audio files\n",
" - ### assumes that mp4s have been converted to mp3s\n",
"\n",
"### quick and dirty way, in a directory with mp4,\n",
" - ### `mkdir audio; for f in *.mp4; ffmpeg -i $f audio/${f:r}.mp3`"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d59ddb47",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 194,
"id": "d81ee3d2",
"metadata": {},
"outputs": [],
"source": [
"\n",
"def make_ordered_unique(l):\n",
" seen = set()\n",
" unique_list = []\n",
" for item in l:\n",
" if item not in seen:\n",
" seen.add(item)\n",
" unique_list.append(item)\n",
" return unique_list \n",
"\n",
"# concatenate audio file into one mp3\n",
"def concat_files(flist, trim_seconds=0):\n",
" arr, (old_sr, old_nc) = load_audio_as_array(flist[0])\n",
" if old_nc == 2:\n",
" arr = arr[:, 0]\n",
" pieces = [arr]\n",
" for f in tqdm.notebook.tqdm(flist[1:]):\n",
" arr, (sample_rate, n_channels) = load_audio_as_array(f)\n",
" if sample_rate != old_sr:\n",
" raise RuntimeError(\"poo\")\n",
" if n_channels == 2:\n",
" arr = arr[:, 0]\n",
" if trim_seconds > 0:\n",
" trim_samples = int(round(trim_seconds * sample_rate))\n",
" arr = arr[:-trim_samples]\n",
" pieces.append(arr)\n",
" return np.concatenate(pieces), sample_rate\n",
"\n",
"\n",
"def write_concatted_mp3(manifest_entry, force=False, trim_seconds=0):\n",
" vid_in_order = [\n",
" l.replace(\"https://vimeo.com/\", \"\") + \".mp3\"\n",
" for l in manifest_entry[\"links\"][\"vimeo\"]\n",
" ]\n",
"\n",
" files_in_order = make_ordered_unique(vid_in_order)\n",
" datestr = manifest_entry[\"url\"].split(\"_\")[-1]\n",
" audio_dir = os.path.join(\"/mnt/data-ssd-1/data/nz-parl/\", datestr, \"audio/\")\n",
" \n",
" wav_fname = os.path.join(audio_dir, \"whole-audio.wav\")\n",
" mp3_fname = os.path.join(audio_dir, \"whole-audio.mp3\")\n",
" \n",
" if os.path.exists(mp3_fname) and not force:\n",
" print(f\"file {mp3_fname} exists. Skipping...\")\n",
" return mp3_fname\n",
" \n",
" files_in_order = [\n",
" os.path.join(audio_dir, l)\n",
" for l in files_in_order\n",
" ]\n",
" files_in_order = list(reversed(files_in_order))\n",
" \n",
" whole_array, sample_rate = concat_files(files_in_order, trim_seconds=trim_seconds)\n",
" byte_width = whole_array.dtype.itemsize\n",
" print(f\"writing wav to {wav_fname}\")\n",
" with open(wav_fname, \"wb\") as fi:\n",
" write_wav(fi, whole_array.tobytes(), byte_width=byte_width, sample_rate=sample_rate)\n",
" \n",
" \n",
" convert_audio(wav_fname, mp3_fname)\n",
" os.remove(wav_fname)\n",
" return mp3_fname"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "4bdbd2d1",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 195,
"id": "079a01e3",
"metadata": {
"scrolled": false
},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "584888abc97544aaa21d2bfaeec0754d",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
" 0%| | 0/25 [00:00\"\n",
"SPEAKER_MODIFIER = \"speaker_modifier\"\n",
"\n",
"\n",
"def parse_section_better(section):\n",
"\n",
" section_class = section.get(\"class\")\n",
" parsed = []\n",
" for child in section.children:\n",
" if isinstance(child, bs4.element.Tag):\n",
" if child.name == \"strong\":\n",
" d = {\"_type\": STRONG, \"name\": child.get(\"name\", UNK_NAME)}\n",
" else:\n",
" d = {\"_type\": \"tag\", \"name\": child.get(\"name\", UNK_NAME)}\n",
" elif isinstance(child, bs4.element.NavigableString):\n",
" d = {\"_type\": \"string\", \"name\": \"text\"}\n",
" else:\n",
" raise\n",
"\n",
" d[\"text\"] = child.text\n",
" if d[\"text\"].startswith(\":\"): # bold speaker tag doesn't include colon\n",
" d[\"text\"] = d[\"text\"].lstrip(\":\").strip()\n",
" d[\"section_class\"] = section_class\n",
" parsed.append(d)\n",
"\n",
" for item_prev, item, item_next in zip(parsed[:-1], parsed[1:-1], parsed[2:]):\n",
" if (\n",
" item[\"text\"].strip().lower() == \"to the\"\n",
" and item_prev[\"_type\"] == STRONG\n",
" and item_next[\"_type\"] == STRONG\n",
" ):\n",
" item[\"_type\"] = SPEAKER_MODIFIER\n",
" return parsed\n",
"\n",
"\n",
"def parse_transcript_better(soup):\n",
" main_body = soup.find_all(\"body\")[0]\n",
" parsed = []\n",
" for section in main_body.find_all(\"p\"):\n",
" parsed += parse_section_better(section)\n",
" return parsed"
]
},
{
"cell_type": "code",
"execution_count": 198,
"id": "88db913d",
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "0346393cbe164f5b98e908463df2a3f2",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
" 0%| | 0/25 [00:00 0 and os.path.exists(slice_meta_fname) and not force:\n",
" print(f\"found {num_slice_wavs} existing slices in {output_dir}, skipping...\")\n",
" with open(slice_meta_fname, \"r\") as fi:\n",
" return json.load(fi)\n",
" \n",
" # if not skip, make slices and write them\n",
" slices_audio, slices_meta, full_text, additional_info = make_segments(data_dir)\n",
"\n",
" output_slice_meta = []\n",
" for audio_slice, meta in tqdm.notebook.tqdm(\n",
" zip(slices_audio, slices_meta), total=len(slices_meta)\n",
" ):\n",
" interval, other = meta\n",
" if not other:\n",
" continue\n",
" st, ed = interval\n",
" txt = other[0][1]\n",
" fn = f\"{st}_to_{ed}\".replace(\".\", \"\") + \".wav\"\n",
" fname = os.path.join(output_dir, fn)\n",
" write_wav(fname, audio_slice)\n",
" output_slice_meta.append({\"filename\": fname, \"interval\": list(interval), \"text\": txt})\n",
" with open(slice_meta_fname, \"w\") as fi:\n",
" json.dump(output_slice_meta, fi)\n",
" return output_slice_meta"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6b5e1089",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 204,
"id": "ae4f06a8",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"['/mnt/data-ssd-1/data/nz-parl/20220317',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220316',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220315',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220310',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220309',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220308',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220303',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220302',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220301',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220217',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220215',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220210',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220209',\n",
" '/mnt/data-ssd-1/data/nz-parl/20220208',\n",
" '/mnt/data-ssd-1/data/nz-parl/20211215',\n",
" '/mnt/data-ssd-1/data/nz-parl/20211214',\n",
" '/mnt/data-ssd-1/data/nz-parl/20211209',\n",
" '/mnt/data-ssd-1/data/nz-parl/20211208',\n",
" '/mnt/data-ssd-1/data/nz-parl/20211207',\n",
" '/mnt/data-ssd-1/data/nz-parl/20211125',\n",
" '/mnt/data-ssd-1/data/nz-parl/20211123',\n",
" '/mnt/data-ssd-1/data/nz-parl/20211117',\n",
" '/mnt/data-ssd-1/data/nz-parl/20211116',\n",
" '/mnt/data-ssd-1/data/nz-parl/20211111',\n",
" '/mnt/data-ssd-1/data/nz-parl/20211110']"
]
},
"execution_count": 204,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"data_dirs = [m[\"mp3_filename\"] for m in manifest]\n",
"data_dirs = [os.path.split(os.path.split(d)[0])[0] for d in data_dirs]\n",
"data_dirs"
]
},
{
"cell_type": "code",
"execution_count": 205,
"id": "c9927315",
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "04a22147650743eca5642a14090ca254",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
" 0%| | 0/25 [00:00\n",
" \n",
" Your browser does not support the audio element.\n",
" \n",
" "
],
"text/plain": [
""
]
},
"execution_count": 209,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"ii = 188\n",
"print(all_slice_meta[ii][\"text\"])\n",
"Audio(all_slice_meta[ii][\"filename\"])\n"
]
},
{
"cell_type": "code",
"execution_count": 190,
"id": "2a12e379",
"metadata": {},
"outputs": [],
"source": [
"import sox"
]
},
{
"cell_type": "code",
"execution_count": 101,
"id": "34c4dba2",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"1"
]
},
"execution_count": 101,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"1"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e0d75f04",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 20,
"id": "55c6a6c3",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'filename': '/mnt/data-ssd-1/data/nz-parl/processed_data/slices/20211207/2472144_to_2472981.wav',\n",
" 'interval': [24721.44, 24729.81],\n",
" 'text': 'Thank you'}"
]
},
"execution_count": 20,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"all_slice_meta[ii]"
]
},
{
"cell_type": "markdown",
"id": "d3129cd6",
"metadata": {},
"source": [
"# bad files\n",
" - /mnt/data-ssd-1/data/nz-parl/processed_data/slices/20211207/2472144_to_2472981.wav"
]
},
{
"cell_type": "code",
"execution_count": 107,
"id": "d0a4323c",
"metadata": {},
"outputs": [],
"source": [
"# for d in glob.glob(base_dir + \"*\"):\n",
"# if not os.path.isdir(d):\n",
"# continue\n",
"# audio_dir = os.path.join(d, \"audio\") \n",
"# print(audio_dir)\n",
"# #os.makedirs(audio_dir, exist_ok=True)\n",
"# mp4_files = glob.glob(os.path.join(d, \"*.mp4\"))\n",
"# if not glob.glob(os.path.join(audio_dir, \"*.mp3\")):\n",
"# for mp4_file in mp4_files:\n",
" "
]
},
{
"cell_type": "code",
"execution_count": 31,
"id": "83c29eff",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "15a497ed",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "6bdbbd98",
"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
}