{
"cells": [
{
"cell_type": "code",
"execution_count": 97,
"id": "20a5b246",
"metadata": {},
"outputs": [],
"source": [
"import json\n",
"import random\n",
"\n",
"from suno_utils.utils.notebook import Audio, transcribe"
]
},
{
"cell_type": "code",
"execution_count": 98,
"id": "edc9b186",
"metadata": {},
"outputs": [],
"source": [
"import string\n",
"import unidecode\n",
"from suno_utils.utils.text import normalize_whitespace\n",
"\n",
"allowed_chars = set(string.ascii_lowercase + string.digits + \" '\")\n",
"\n",
"def _normalize_text_for_asr(text):\n",
" text = unidecode.unidecode(text)\n",
" text = \"\".join([c if c in allowed_chars else \" \" for c in text.lower()])\n",
" text = normalize_whitespace(text)\n",
" return text\n",
"\n",
"def _normalize_nemo_pred(text):\n",
" text = \" \".join([w for w in text.split() if w not in (\"uh\", \"um\")])\n",
" text = text.replace(\"-\", \" \")\n",
" text = normalize_whitespace(text)\n",
" return text\n",
"\n",
"from nemo.collections.asr.metrics.wer import word_error_rate\n",
"\n",
"def _get_cer(a, b):\n",
" a = normalize_whitespace(a)\n",
" b = normalize_whitespace(b)\n",
" if a == b:\n",
" return 0\n",
" cer = word_error_rate([a], [b], use_cer=True)\n",
" return cer"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0b6a25eb",
"metadata": {},
"outputs": [],
"source": [
"to_data_dir = \"/mnt/data-ssd-1/data/nz-parl_georg/\"\n",
"\n",
"with open(to_data_dir + \"debug_info.json\") as f:\n",
" debug_info = json.load(f)\n",
" \n",
"with open(to_data_dir + \"slices_info.json\") as f:\n",
" slices_info = json.load(f)\n",
"print(len(slices_info), \"events available\")\n",
"\n",
"# assemble data for all parsed slices\n",
"slice_data = []\n",
"for uid, v in slices_info.items():\n",
" for n, (_, turns) in enumerate(v):\n",
" filepath = to_data_dir + \"slices/{}/{}.wav\".format(uid, n)\n",
" if filepath not in fp2text:\n",
" continue\n",
" norm_text = _normalize_text_for_asr(\" \".join([e[1] for e in turns]))\n",
" slice_data.append((uid, filepath, turns, norm_text))\n",
" \n",
"# make transcripts\n",
"slice_filepaths = [e[1] for e in slice_data]\n",
"out = transcribe(slice_filepaths)\n",
"fp2text = {fp: _normalize_nemo_pred(t) for fp, t in zip(slice_filepaths, out)}\n",
"\n",
"# add to slice data\n",
"tmp_slice_data = []\n",
"for uid, slice_filepath, turns, text in slice_data:\n",
" transcript = fp2text[slice_filepath]\n",
" tmp_slice_data.append((uid, slice_filepath, turns, text, transcript))\n",
"slice_data = tmp_slice_data"
]
},
{
"cell_type": "code",
"execution_count": 88,
"id": "d8bb2ed8",
"metadata": {
"scrolled": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"46210\n",
"events that happened in aotearoa new zealand in the 2000s and of course in 2019 because the reality is that we now know this country is not immune from terrorist activity and the threat of terrorism\n",
"\n",
"events that happened in alteldoor new zealand in the two thousand and of course in two thousand and nineteen because the reality is we now know this country is not immune from terrorist activity and the threat of terrorism\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"47776\n",
"been identified in managed isolation and quarantine miq among those who had completed a pre departure test before arriving in new zealand\n",
"\n",
"we been identified in miq among those who had completed a pre departure test before arriving in new zealand\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"35528\n",
"and unable to be influenced through their remuneration to be partial to a government decision the last of the officers to whom this bill relates are the chairperson and deputy chairpersons of the human rights\n",
"\n",
"and unable to be influenced through their reenumeration to be partial to a government decision madam skdent the last of the officers to whom this bill relates are the chairperson and deputy chairpersons of the human rights\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"20259\n",
"to acknowledge that yesterday was a big day for the work to eliminate family violence and sexual violence for so many over so long many many decades who have been\n",
"\n",
"to acknowledge that yesterday was a big day for the work to eliminate family violence and sexual violence for so many over so long many many decades who have been\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"32109\n",
"interject during an answer they do invite the minister to go on in a way for longer than would normally be allowed and that's certainly occurred on that occasion i would have cut the minister off\n",
"\n",
"interject during an answer they do invite the minister to go on in a way for longer than would normally be allowed and that certainly occurred on that occasion i would have cut the minister off\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"45851\n",
"which can be addressed and which the minister himself has agreed to look into as we move forward so i would like to\n",
"\n",
"which can be addressed and which the minister himself has agreed to look into as we move forward so i would like to\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"37106\n",
"motorcycle or anything flash or with a hundred thousand dollars' worth of jewellery on them that that can be taken unless\n",
"\n",
"motorcycle or anything flash or thousands of dollars hundreds of thousand dollars worth of jewellery on them that that can be taken unless\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"31602\n",
"going off to a festival and taking something when you've instilled in them since they were a child don't take things from strangers or\n",
"\n",
"going off to a festival and taking something when you've instilled in them since they were a child don't take things from strangers or\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"33201\n",
"universal human rights and that declaration begins that all people it actually said all men but i'm sure they meant all people are born free and equal this motion\n",
"\n",
"universal human rights and that declaration begins that all people they actually said all men but i am sure they meant all people are born free and equal this motion\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"2967\n",
"box of weet bix in the morning it's just incredible how much money people are needing to spend every week at the supermarket and that's on top of the cost of petrol going up\n",
"\n",
"box of wheat bits in the morning 's beside you know it's just incredible how much money people are needing to spend every week at the supermarket and that's on top of the cost of petrol going up\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"# listen to random sample\n",
"for _ in range(10):\n",
" n = random.choice(list(range(len(slice_data))))\n",
" uid, slice_filepath, turns, text, transcript = slice_data[n]\n",
" print(n)\n",
" print(text)\n",
" print()\n",
" print(transcript)\n",
" a = Audio.from_file(slice_filepath)\n",
" a.play()"
]
},
{
"cell_type": "code",
"execution_count": 89,
"id": "83070f21",
"metadata": {},
"outputs": [],
"source": [
"# get wer and sort by difficulty\n",
"cers = [_get_cer(e[-2], e[-1]) for e in slice_data]\n",
"len_text = [len(e[-2]) for e in slice_data]\n",
"len_transcript = [len(e[-1]) for e in slice_data]"
]
},
{
"cell_type": "code",
"execution_count": 90,
"id": "506e383a",
"metadata": {},
"outputs": [],
"source": [
"import pandas as pd\n",
"\n",
"df = pd.DataFrame({\n",
" \"idx\": [n for n in range(len(slice_data))],\n",
" \"filepath\": [e[1] for e in slice_data],\n",
" \"text\": [e[-2] for e in slice_data],\n",
" \"transcript\": [e[-1] for e in slice_data],\n",
" \"cer\": cers,\n",
" \"len_text\": len_text,\n",
" \"len_transcript\": len_transcript,\n",
"})"
]
},
{
"cell_type": "code",
"execution_count": 91,
"id": "731c1104",
"metadata": {},
"outputs": [],
"source": [
"sub_df = df[\n",
" (df[\"len_text\"] >= 50) &\n",
" (df[\"len_transcript\"] >= 50) & \n",
" (df[\"len_text\"] / df[\"len_transcript\"] >= 0.8) & \n",
" (df[\"len_text\"] / df[\"len_transcript\"] <= 1.2)\n",
"].sort_values(\"cer\", ascending=False)"
]
},
{
"cell_type": "code",
"execution_count": 92,
"id": "72a42579",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"30229 - 0.47\n",
"yeah we're supportive of quick changes here mr speaker my you've changed my apologies but\n",
"\n",
"yeah we we're supportiveis speaking myy you have changed my apologiesum but\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"45808 - 0.47\n",
"all about so when i talk about camping now interruption look i'm\n",
"\n",
"is all about so so we when i talk about camping now like the\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"26434 - 0.46\n",
"and it's acknowledged that there's 1 47 billion and 90 percent of that which is 1 354 billion is over four expense\n",
"\n",
"it is acknowledged that is one forty one four seven billion ninety of that which is one three five four billion is over four expen\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"190 - 0.44\n",
"well speaking to that point of order mr speaker mr speaker i think if the member chose\n",
"\n",
"honourable grand robert mister speaker mister speaker i think if the member chose\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"25792 - 0.44\n",
"disability support services that the ministry of social development msd does fund i have to say that when we came into government in 2017 for nine years just about all of the social\n",
"\n",
"disability support services that msd does fund i have to say that when we came into government in two thousand and two thousand and seventeen for nine years just about all of the social sector\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"9584 - 0.44\n",
"and that he was not contemplating using rapid antigen tests rats in aotearoa this statement was made while rats were\n",
"\n",
"and that he was not contemplating using rats in altero this statement was made at the same rats where being deploined well\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"47237 - 0.44\n",
"in fact if we look at the benefit assistance numbers now\n",
"\n",
"and in fact if we look at the unemployment numbers near six\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"45580 - 0.44\n",
"tell us about it well fun ho toys is inglewood's which is a fabulous it was the sort of 1930s 40s and 50s when it was huge those metal toys\n",
"\n",
"well fuo toys a inglewoods which is is a fabulous it was it was the sort of the nineteen thirties forties and fifties when it was huge those metal toy\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"3516 - 0.44\n",
"of course this also under proposed new section 141ee there we are i'm just\n",
"\n",
"of course this also under clause one for one double e there are there we are under\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"25131 - 0.44\n",
"will she halt the government's 100 million infrastructure reference group shovel ready investment in the pumped hydro feasibility study\n",
"\n",
"to the minister will she hold the government's one hundred million irg shovelready investment in the pump hydrofeasibility study\n"
]
},
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"for _, row in sub_df.head(20).tail(10).iterrows():\n",
" print(row[\"idx\"], \"-\", round(row[\"cer\"], 2))\n",
" print(row[\"text\"])\n",
" print()\n",
" print(row[\"transcript\"])\n",
" a = Audio.from_file(row[\"filepath\"])\n",
" a.play()"
]
},
{
"cell_type": "code",
"execution_count": 93,
"id": "f7b27f41",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"('20211117',\n",
" '/mnt/data-ssd-1/data/nz-parl_georg/slices/20211117/824.wav',\n",
" [['Carmel Sepuloni',\n",
" 'disability support services that the Ministry of Social Development (MSD) does fund, I have to say that when we came into Government in 2017, for nine years just about all of the social']],\n",
" 'disability support services that the ministry of social development msd does fund i have to say that when we came into government in 2017 for nine years just about all of the social',\n",
" 'disability support services that msd does fund i have to say that when we came into government in two thousand and two thousand and seventeen for nine years just about all of the social sector')"
]
},
"execution_count": 93,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"slice_data[25792]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8cb7e8ce",
"metadata": {},
"outputs": [],
"source": [
"# interruption - transcribed\n",
"# multiple members interject - transcribed\n",
"# the Ministry of Social Development (MSD) -> pronounced \"msd\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "5b47f498",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "5981e163",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 57,
"id": "96a29120",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"Audio(14.8 secs, 16000 Hz, 16 bits, 1 ch)"
]
},
"execution_count": 57,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"a = Audio.from_file(\"/mnt/data-ssd-1/data/nz-parl_georg/slices/20220317/0.wav\")\n",
"a"
]
},
{
"cell_type": "code",
"execution_count": 59,
"id": "39cf9232",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"a.play()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "25bdfd0a",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "19667913",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "2dc2dd9b",
"metadata": {},
"outputs": [],
"source": [
" for n, ((from_s, to_s), _) in enumerate(slices_meta):\n",
" to_wav_filepath = to_slice_dir + \"{}.wav\".format(n)\n",
" if not os.path.exists(to_wav_filepath):\n",
" audio_16khz.get_slice(from_s=from_s, to_s=to_s).write_wav(to_wav_filepath)"
]
},
{
"cell_type": "code",
"execution_count": 51,
"id": "59ab3ebd",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[[13932.1, 13939.81],\n",
" [['Chris Hipkins',\n",
" 'the Financial Sector (Climate-related Disclosures and Other Matters) Amendment Bill, and the COVID-19 Response (Management Measures) Legislation Bill;']]]"
]
},
"execution_count": 51,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"slices_info[\"20210930\"][0]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "aaf04b71",
"metadata": {},
"outputs": [],
"source": [
"# TODO: play stretches of non-covered audio and look for hard bits\n",
"# TODO: listen to random slices\n",
"# TODO: check low slice coverage fractions:\n",
"# ('20210930', 61.4, 29.6),\n",
"# ('20211028', 76.8, 35.1),\n",
"# ('20211209', 79.8, 37.7),\n",
"# ('20211215', 77.0, 39.2),\n",
"\n",
"# TODO: super slow\n",
"# ('20210921', 28.1),\n",
"# ('20211117', 28.9),\n",
"# ('20211110', 45.8),\n",
"# ('20211020', 49.6)]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ad6a2fb2",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "4822a163",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "15ff7a5d",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "20e8a1d0",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "d78c7742",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "e9735db1",
"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
}