bolAIndia

Human-side speech from production call recordings, cut into utterance-level chunks by a two-engine VAD (Silero + TEN) and transcribed by third-party ASR providers. Each row keeps the transcript, the provider's confidence, and full provenance back to the source recording.

Alongside it, open Indian-language speech corpora converted to the same schema (16 kHz mono FLAC, one utterance per row), each in a config of its own and tagged with where it came from.

Size

	hours	chunks	configs
Call-centre speech, vendor transcripts	20.56K	24.30M	3
Open Indian-language datasets	129.30K	47.30M	49
All	149.86K	71.60M	52

K = thousand, M = million. Exact figures are in STATS.md.

Repeats are removed: an open-dataset clip is published once, and call-centre rows that repeated another row's audio and transcript were removed on 5 October 2026. A call chunk that two vendors transcribed differently keeps one row per vendor.

Sources

One config per transcription system, so their output stays separable.

config (source_id)	provider	model	hours	chunks	shards
vendor-a	vendor-a	undisclosed	418	478K	42
combined	all vendors	mixed	20.1K	23.8M	4.6K
vendor-b	vendor-b	undisclosed	10	12.2K	1

The breakdowns below round their figures, except the open datasets' hours; STATS.md has every figure in full.

Who transcribed what

The combined config interleaves several transcription systems within each shard, so this is the breakdown across the whole dataset. The systems differ substantially, so treat them as separate sources when training.

Each row carries the system that produced it in its provider and model columns; the labels below are withheld aliases for the same systems, in the same order of size.

vendor	hours	chunks	share of hours
vendor-a	17.4K	20.5M	84.7%
vendor-b	3K	3.6M	14.4%
vendor-c	162	187K	0.8%
vendor-d	28.4	31.7K	0.1%
combined	5.5	6.5K	<0.1%

Open Indian-language datasets

Each open dataset is a config of its own, and every row names its source in source_id (the config) and source_db (the original repo or URL).

No clip is published twice. Across every open dataset, a row is left out if an earlier row already carries the same audio (identical 16 kHz samples), the same chunk_id, or the same broadcast clip, so datasets that repeat themselves (a bulletin ingested twice, a segment copied into the next file) keep one copy. The same speech recorded or encoded differently is not caught.

Hours here are exact. Held out is the part tagged as an evaluation split (see below); duplicates removed counts rows left out because another row already carried them.

config (source_id)	dataset	transcripts	languages	hours	of which held out	chunks	shards	duplicates removed
yodas3	YODAS3	youtube asr captions	10: Hindi, Punjabi, Bengali …	66,650.61	0.00	22,725,807	7,387	46,082
newsonair	NewsOnAIR preserved audio	website scripts where verified; otherwise untranscribed	6: Hindi, English, Telugu …	12,228.33	0.00	1,600,667	1,791	176
indicvoices	IndicVoices	human	22: Hindi, Tamil, Assamese …	11,855.81	125.96	6,222,099	1,346	24
nptel	NPTEL	human	10: English, Tamil, Hindi …	7,261.47	0.00	3,073,312	754	3,325
shrutilipi	Shrutilipi	aligned bulletin text	16: Hindi, Marathi, Tamil …	3,926.75	0.00	2,206,066	480	19,731
numo-indic-speech	Numo Indic Speech	read prompts	5: Telugu, Hindi, Bengali …	3,401.41	0.00	330,338	434	0
chaashini	Chaashini	asr	33: Hindi, English, Urdu …	2,932.81	0.00	1,382,550	409	15,435
spring-inx-languages	SPRING-INX (nine more languages)	human	9: Bengali, Marathi, Punjabi …	2,315.74	296.42	987,900	301	152,671
worldspeech	WorldSpeech	asr	16: Hindi, Tamil, Marathi …	1,953.75	88.81	510,473	215	2,793
vaani	Vaani	human	61: Hindi, Telugu, Bengali …	1,928.26	386.61	1,317,716	214	2,523
pretraining-v1-cv22-sidon	Common Voice 22, restored (Pretraining-V1)	read prompts	13: Bengali, Tamil, Urdu …	1,804.21	0.00	1,425,647	230	12,519
podcast-captions	Indic podcasts with publisher transcripts	asr captions	8: Hindi, Tamil, Telugu …	1,635.60	0.00	325,425	219	3,092
pretraining-v1-kathbath	Kathbath (Pretraining-V1)	read prompts	12: Tamil, Marathi, Kannada …	1,475.20	0.00	805,721	207	0
open-bible	OpenBibleTTS (South Asian languages)	aligned verse text	13: Nepali, Kannada, Assamese …	1,236.72	0.00	395,695	168	0
rasa	Rasa	read prompts	22: Bengali, Punjabi, Assamese …	1,129.36	113.65	634,166	128	2,506
pretraining-v1-syspin	SYSPIN (Pretraining-V1)	read prompts	9: Magahi, Chhattisgarhi, Bengali …	1,080.52	12.18	511,224	151	275,401
indic-tts-966h	Indic-TTS-966h	not documented	7: Marathi, Malayalam, Tamil …	965.42	0.00	159,791	122	90
madasr-bengali	MADASR 2023 Bengali (RESPIN)	read prompts	Bengali	537.91	0.00	371,958	64	99
pretraining-v1-indictts	IndicTTS (Pretraining-V1)	read prompts	13: English, Gujarati, Assamese …	502.07	0.00	282,243	64	11,765
snow-mountain	Snow Mountain	aligned verse text	13: Kannada, Malayalam, Hindi …	437.67	98.06	153,632	56	1
uwc-hindi	ASR-UWC Hindi ASR corpus	not documented	Hindi	429.42	0.00	266,134	47	23,175
mann-ki-baat	Mann Ki Baat (BhasaAnuvaad)	aligned official text	12: Marathi, Tamil, English …	404.11	0.00	219,278	51	7
iisc-mile-kannada	IISc-MILE Kannada ASR corpus	read prompts	Kannada	335.06	72.44	101,814	43	3,723
bbc-podcasts	BBC Indian-language podcasts (untranscribed)	untranscribed	5: Hindi, English, Tamil …	250.35	0.00	57,276	38	610
subak-ko	SUBAK.KO	human	Bengali	240.91	40.80	77,552	34	2
spring-inx	SPRING-INX Hindi	human	Hindi	234.69	47.32	91,926	33	114
bhojpuri-rural	Rural Women Bhojpuri	human	Bhojpuri, Hindi	202.69	1.26	78,805	23	0
openslr-53-bengali	Google crowdsourced Bengali (OpenSLR 53)	read prompts	Bengali	195.86	0.00	198,788	24	0
aawaz	aawaz audio shows (untranscribed)	untranscribed	4: Hindi, English, Marathi …	195.18	0.00	66,086	24	2,660
omnilingual-asr-corpus	Omnilingual ASR Corpus (South Asian languages)	human	36: Garhwali, Korwa, Konkani (individual language) …	185.95	53.99	17,821	25	329
fleurs	FLEURS	read prompts	14: Marathi, Sindhi, Malayalam …	177.53	53.29	51,938	20	0
rural-women-asr-v2	Rural Women ASR v2	human	Bhojpuri, Hindi	158.31	0.00	61,437	24	246
iisc-mile-tamil	IISc-MILE Tamil ASR corpus	read prompts	Tamil	150.08	16.81	89,390	21	10
pretraining-v1-msft-indian	Microsoft Indian-language speech (Pretraining-V1)	read prompts	Gujarati, Tamil, Telugu	134.92	0.00	115,392	18	0
gurbani-sehajpath	Gurbani Sehaj Path	captions corrected to scripture	Punjabi	123.99	0.00	49,274	19	0
mucs-hinglish	MUCS 2021 Hindi-English code-switched	human	Hindi	102.62	11.19	57,039	14	0
pretraining-v1-spicor	SPICOR Indian English (Pretraining-V1)	read prompts	English	99.34	1.12	50,423	15	0
vanipedia	Vanipedia lectures (BhasaAnuvaad)	aligned official text	English	93.35	0.00	64,630	10	53
svq	Simple Voice Questions	read prompts	10: Bengali, Telugu, Urdu …	82.72	82.72	53,375	9	0
all-hindi-asr	All Hindi ASR v1.1 (filtered)	not documented	Hindi	66.38	0.00	27,393	9	14,246
indic-context-eval	IndicContextEval	human	8: Telugu, Marathi, Hindi …	55.92	55.92	16,874	7	2
imasc	IMaSC	read prompts	Malayalam	49.57	0.00	34,438	8	30
pretraining-v1-synthetic-v1	synthetic_v1 (Pretraining-V1)	synthetic	9: Hindi, Bengali, Kannada …	21.04	0.00	9,941	3	0
gaps-nptel	NPTEL lectures (GAPS)	human	English	15.30	0.00	6,465	4	4
lahaja	Lahaja	human	Hindi	12.32	12.32	6,152	2	0
navana-hindi	Navana Hindi ASR Benchmark	human	Hindi	11.16	11.16	7,649	2	0
pretraining-v1-ivr	IndicVoices-R (Pretraining-V1)	human	22: Dogri, Maithili, Telugu …	6.49	0.00	2,094	8	653,567
pretraining-v1-elise-hindi	Elise Hindi (Pretraining-V1)	not documented	Hindi	2.36	0.00	1,147	1	0
pretraining-v1-multilingual-tts	multilingual_tts (Pretraining-V1)	synthetic	Bengali, Hindi, Urdu	1.95	0.00	991	2	0
all open datasets				129,299.20	1,582.02	47,303,952	15,278	1,247,011

• yodas3: Whole YouTube videos in 10 Indian languages, cut into utterances on the words of their auto-generated (ASR) captions: noisy, machine-made labels. Uploader subtitle tracks (translations, lyrics) are not what was said and are left out; so are chunks not mostly in the language's own script.
• newsonair: 70,809 rows (187.1 h) carry All India Radio's own bulletin scripts, matched to the recording and checked against the audio word by word; the recording is cut at its pauses around them. The other rows are untranscribed: exclude them from supervised ASR. Alignment is automated, not human-reviewed. Original recordings and scripts remain archived.
• indicvoices: Crowd-recorded read, extempore and conversational speech from across India, transcribed and verified by AI4Bharat. The text is the verbatim transcript (what was said, not the standardised text); rows marking part of the audio unintelligible are left out. Its valid split (the IndicVoices benchmark) is tagged valid.
• nptel: NPTEL university lectures from AI4Bharat's BhasaAnuvaad: the original lectures in Indian English (en) and voice-over dubs in 9 Indian languages, whose text is the translated script the artist read. Kept only where the text matches an ASR pass (alignment >= 0.8, stored in confidence) and the clip is 0.5-30 s. Numbers are written as digits.
• shrutilipi: All India Radio news read against published bulletin text, aligned automatically.
• numo-indic-speech: Conversational scripts read by speakers in India and Bangladesh, in Bengali, Hindi, Tamil and Telugu (Poseidon AI). confidence is 1 minus the dataset's own character error rate check; its WER and CER, its synthetic-voice suspicion score, the topic and the speaker's age, gender, country and region are in vad_detail. Clips over 60 s are left out. The Bengali part is also published as the evaluation set of psdn-ai/bangla-10k.
• chaashini: Kapture CX's quality-gated corpus of clean single-speaker speech in 33 Indian languages and Indian English: clips cut at pauses from public spoken-word YouTube recordings (talks, interviews, narration, lectures, podcasts). The filtering is Chaashini's: recordings dominated by music or singing dropped, speech activity detection, one speaker per clip (diarised, overlaps removed), DNSMOS, SNR and music/noise scores within strict limits (borderline clips enhanced and re-scored, flagged enhanced), and ASR transcripts kept only with plausible character rates and confident recognition, so the text is that ASR's output. Rows keep Chaashini's clip id, source recording, speaker and segment index; confidence is its ASR confidence, language_confidence and speech_ratio are its own, and its other scores are in vad_detail under chaashini, so its filters still apply (e.g. dnsmos_ovrl >= 3.2 and enhanced == false). Its source videos whose transcripts are largely found in the yodas3 config (compared by content: yodas3's video ids are anonymised) are left out as duplicates. Cite: Kapture CX, "Chaashini: a quality-gated multilingual Indian speech corpus" (2026).
• spring-inx-languages: IIT Madras's SPRING-INX R1 and R2 releases (NLTM) for Assamese, Bengali, Gujarati, Kannada, Malayalam, Marathi, Odia, Punjabi and Tamil, from SPRING Lab's Hugging Face mirrors (SPRINGLab/SPRING_INX__<R1|R2>): phone conversations and read speech, transcribed and checked by hand. The Hindi releases are the spring-inx config. Their train/validation/test splits are kept; config_id names the language and release. Audio R2 re-released from R1 is taken once.
• worldspeech: Every Indian-language config: All India Radio news in 16 languages, Hindi legislatures and Mann Ki Baat, and the Kerala, Maharashtra and Punjab assemblies, plus Bengali (Bangladesh), Nepali (Nepal), Tamil (Sri Lanka) and Urdu (Pakistan); the config is in config_id. The text is WorldSpeech's asr_transcript (machine output, within CER 0.3 of the official record); confidence is 1 - CER.
• vaani: The transcribed part of IISc/ARTPARK's Vaani: people in 165 districts describing an image, in 64 languages and dialects. Annotation markup is stripped the way the Vaani benchmark scores, and rows marking audio unintelligible are left out. Its validation and test splits keep those tags; every segment of a recording Vaani-Benchmark-V1.0 took segments from is tagged benchmark.
• pretraining-v1-cv22-sidon: Common Voice 22 clips cleaned by the Sidon speech-restoration model. Only the rows in Indian languages are taken (13, most of them Bengali, Tamil and Urdu); its English, Pashto, Dhivehi and Saraiki are not.
• podcast-captions: Episodes of Indian-language podcasts (found from feedspot's lists by bolai discover podcasts) whose feeds attach a timestamped transcript (Podcasting 2.0 podcast:transcript, SRT or VTT): nearly all are Spotify's automatic transcripts, so this is machine-captioned speech, like yodas3. Episodes are cut at the transcript's cues (merged up to 25 s, never inside a cue, never overlapping); rows whose text is not mostly in the language's own script, or whose speaking rate is implausible for their length (cues drifted off the audio), are left out. Episodes found in the chaashini or yodas3 configs by shared distinctive phrases, and episodes published twice, are left out (podcast_captions_exclude.json). recording_id is the episode, config_id the show; show, episode title, date and transcript URL are in vad_detail.
• pretraining-v1-kathbath: AI4Bharat's read speech in 12 Indian languages from speakers across India. Its own valid/test splits are not marked in this copy, so every row is tagged train.
• open-bible: Bible readings from Open.Bible, cut at verses and paired with the verse text, in 13 South Asian languages (Assamese, Bengali, Chhattisgarhi, Gujarati, Hindi, Kannada, Malayalam, Marathi, Nepali, Punjabi, Tamil, Telugu, Urdu); the dataset's other languages are not taken. Book, chapter, verse and testament are in vad_detail. Its speaker labels (SPEAKER_01) are diarisation clusters, scoped by language (Hindi/SPEAKER_01).
• rasa: Expressive studio speech recorded for TTS: the text is the prompt the speaker read.
• pretraining-v1-syspin: IISc SYSPIN TTS recordings: studio speech of a few speakers each in Hindi, Bengali, Marathi, Kannada, Telugu, Maithili, Magahi, Chhattisgarhi and Bhojpuri; rows named EVAL keep that tag.
• indic-tts-966h: About 966 hours of talks in Bengali, Malayalam, Marathi, Punjabi, Tamil and Telugu: people telling their own stories, with English words kept in Latin script. The clips are 24 kHz, processed by vocal isolation, de-reverberation, denoising and enhancement for F5-TTS training. The dataset does not say where the recordings or transcripts come from; the clips themselves point to Josh Talks videos on YouTube (their outros and app adverts are in them), and the text reads as machine transcription ("Josh Talks" comes out as "Jo Stocks"). Clips with no letter of their language's script are tagged en. Talks also found in other configs (compared by content: the enhanced audio cannot match by hash) are left out as duplicates. recording_id is the talk, chunk_index the clip's position in it.
• madasr-bengali: Bengali dialect read speech from the MADASR 2023 challenge (IISc SPIRE Lab's RESPIN), the madasr split of SKNahin/open-large-bengali-asr-data. Its wav2vec2 check (wer, words per second, is_better) is in vad_detail, not used to filter.
• pretraining-v1-indictts: IIT Madras TTS recordings: studio speech of a few speakers each in 13 Indian languages, and English read by Indian speakers.
• snow-mountain: Studio Bible readings cut at verses (Bridge Connectivity Solutions), the authors' cleaned release: the New Testament in Bhadrawahi, Bilaspuri, Dogri, Gaddi, Haryanvi, Kangri, Kulvi, Kulvi Outer Seraji, Mandeali and Pahari Mahasui (most of them in no other open speech set), and more of the Bible in Hindi, Kannada, Malayalam, Tamil and Telugu. Its own train/validation/test lists are kept (test_common is test); verses the authors left out of every list are left out. recording_id is the chapter; book, chapter and verse are in vad_detail. Its Tamil and Telugu are the same recordings as open-bible's (verse durations equal to the millisecond, same text) and are not taken; its Hindi, Kannada and Malayalam are other translations, read separately.
• uwc-hindi: Audio with transcripts, labelled Hindi by its publisher, which does not document where either comes from. Most is Hindi, but some files hold Bhojpuri, Maithili and Marathi as well, tagged hi here like the rest.
• mann-ki-baat: The Prime Minister's Mann Ki Baat radio talks in English and 11 Indian languages, each language's official text aligned to its audio (AI4Bharat BhasaAnuvaad). The text is the spoken language's own transcript, not a translation; confidence is the alignment score. The Hindi part is not taken: the worldspeech config already carries Hindi Mann Ki Baat broadcasts.
• iisc-mile-kannada: IISc MILE Lab's Kannada read speech (OpenSLR 126): 915 speakers on USB microphones in quiet rooms, sentences and word lists; from a Hugging Face mirror of the OpenSLR release. Its train/test split is kept.
• bbc-podcasts: Every episode of BBC's 31 Indian-language and India-focused podcast feeds (Hindi, Marathi, Tamil, Gujarati, Urdu, Nepali, and Indian English), cut into 0.1-30 s clips at speech by the two-engine VAD the transcription pipeline uses. No transcripts exist: source_collection=untranscribed, text is empty, is_unintelligible is null; language_code is the feed's language, unverified per clip. recording_id is the episode; show, title and date are in vad_detail. Filter these rows out of supervised training until transcribed.
• subak-ko: Bangladeshi Bangla speech with human transcripts (SUST CSE); its train/validation/test splits are kept.
• spring-inx: Hindi phone conversations and monologues from IIT Madras (NLTM), R1 and R2 'clean' releases, all splits. R2's re-releases of R1 audio and R1's duplicate recordings are taken once. Every eval utterance is tagged eval: all of an eval recording (under any of its names) and every utterance of the original R1 and R2 test sets, which the clean releases moved into train/dev.
• bhojpuri-rural: Bhojpuri (and some Hindi) from rural women. Only train_real (400 clips) and benchmark (444) are real recordings; train_synthetic, about 99% of the rows, is synthetic speech.
• openslr-53-bengali: Google's crowdsourced Bangladeshi Bengali read speech (OpenSLR 53), the openslr split of SKNahin/open-large-bengali-asr-data; that compilation's other splits are left out (sets bolAIndia has, or of unstated origin). Its wav2vec2 check (wer, words per second, is_better) is in vad_detail, not used to filter.
• aawaz: Every episode of aawaz.com's 634 shows (Hindi, Marathi, Urdu, Gujarati and English: stories, dramas, news, devotional and talk), cut into 0.1-30 s clips at speech by the two-engine VAD the transcription pipeline uses. No transcripts exist: source_collection=untranscribed, text is empty, is_unintelligible is null; language_code is the show's language tag, unverified per clip (some en shows are Hinglish). Music-heavy dramas keep only what the VAD calls speech. Filter these rows out of supervised training until transcribed.
• omnilingual-asr-corpus: Meta's Omnilingual ASR corpus: people answering prompts in their own language, transcribed. Only its 36 South Asian languages are taken, most of them never before in an open ASR set: Angika, Bhili, Bhojpuri, Kanauji, Bundeli, Braj, Bodo, Doteli, Garhwali, Goan Konkani, Hadothi, Jaunsari, Kinnauri, Korwa, Konkani, Maria, Mewari, Nimadi, four Tharu languages, Kota, Sikkimese, Tulu, Kui, Kashmiri, Western Punjabi, Pahari-Potwari, Deccani, two Koli languages, Od, Northern Hindko, Mizo and Kokborok. Its own train/dev/test splits are kept; the prompt id is in vad_detail. Answers run 15-90 s; those over the 60 s clip limit are left out (no timings to cut them by). Speaker labels are per language, so they are scoped by config (bho_Deva/spk10).
• fleurs: Read FLoRes sentences in 14 Indian-language configs (as, bn, gu, hi, kn, ml, mr, ne, or, pa, sd, ta, te, ur), up to three speakers per sentence (config_id names the sentence). Its validation and test splits keep those tags.
• rural-women-asr-v2: AI4Bharat's spontaneous speech of rural women in Hindi and Bhojpuri (extempore answers to prompts), transcribed verbatim; noise and event marks such as are removed from the text. The repo's train-* files repeat the per-language files and are not read. 147 clips re-published from bhojpuri-rural (same transcript and duration, renamed speakers, re-encoded audio) are left out (rural_women_asr_v2_exclude.json). District, state, scenario, gender and age group are in vad_detail.
• iisc-mile-tamil: IISc MILE Lab's Tamil read speech (OpenSLR 127): 531 speakers; from a Hugging Face mirror of the OpenSLR release. Its train/test split is kept.
• pretraining-v1-msft-indian: Read Gujarati, Tamil and Telugu sentences, about 45 hours of each.
• gurbani-sehajpath: Recitations of the Guru Granth Sahib from YouTube, cut at caption cues, with each caption replaced by the matching scripture line (Gurmukhi). Only clips the dataset marks training_usable are taken; its match score and decision are in vad_detail; recording_id is the video.
• mucs-hinglish: The Hindi-English code-switching task of the MUCS 2021 challenge (IIT Madras / Microsoft): Spoken Tutorial lectures, Hindi in Devanagari with English words in Latin script, as one uploader cleaned of stray symbols. Its train, test and blind test are kept (both tests tagged test); the share of English words is in vad_detail.
• pretraining-v1-spicor: English read by Indian speakers (IISc SPICOR project); rows named EVAL keep that tag.
• vanipedia: Lectures in Indian English from Vanipedia with their published transcripts aligned to the audio (AI4Bharat BhasaAnuvaad); confidence is the alignment score.
• svq: Short spoken questions recorded on phones in 12 Indian locales (including Indian English and Bangladeshi Bengali and Pakistani Urdu), each in one of four real conditions: clean, background speech, media, traffic. Published as an evaluation collection, so every row is tagged test.
• all-hindi-asr: Hindi audio with transcripts; the dataset does not document where either comes from.
• indic-context-eval: AI4Bharat's benchmark of domain speech (engineering, medicine, finance and more) in eight languages, dense with technical terms; the transcript without bracketed English glosses is used. An evaluation set: every row is tagged test. Domain, speech style and code-mixing are in vad_detail.
• imasc: ICFOSS Malayalam speech corpus: news sentences read by eight speakers.
• pretraining-v1-synthetic-v1: Synthetic speech (TTS) of conversational lines in eight Indian languages; event tags such as are removed from the text. It labels some Marathi lines Hindi; those are tagged mr (the published label stays in config_id).
• gaps-nptel: NPTEL university lectures in Indian English with NPTEL's own manual transcripts, from GAPS-nptel (an accent-conversion corpus): only its original lecture audio is taken; its two synthetic columns (native-accent and golden-speaker voices) are never read. recording_id is the lecture's YouTube video. 8,953 of its 9,011 lectures are already in the nptel config and are left out (gaps_nptel_exclude.json): only the 61 files holding the other 58 are read, and rows whose transcript marks an unintelligible span, a slide/time reference or an ambiguous [a/b] reading, and the lecture title cards read over music, are left out.
• lahaja: Hindi from 132 speakers in 83 districts, read and extempore, published as a test set only (tagged test).
• navana-hindi: Hindi test clips drawn from CommonVoice, IndicTTS, Kathbath and MUCS (named in config_id) with their human references, tagged test.
• pretraining-v1-ivr: IndicVoices clips enhanced for TTS (48 kHz). The indicvoices config already carries about 98% of them (left out here as duplicates). IndicVoices-R deletes the marks from the transcripts, so clips whose IndicVoices transcript has one are left out as the indicvoices reader leaves out their originals. What remains is clips missing from IndicVoices' release.
• pretraining-v1-elise-hindi: One female voice reading Hindi lines translated from the English Elise set; likely synthetic.
• pretraining-v1-multilingual-tts: Synthetic code-switched speech (TTS); only rows in Indian languages, alone or with English, are taken (about a tenth of it).

NewsOnAIR: same format, transcript availability is explicit

newsonair uses the same Parquet columns and embedded 16 kHz mono FLAC as every other config. Audio awaiting verified transcripts has source_collection=untranscribed, empty text, and null is_unintelligible. Missing transcripts do not mean silence or unintelligible speech. Verified website-script rows use source_collection=website_script_aligned and provider=website_script. A recording whose bulletin script was matched to it and checked against the audio is re-cut at its pauses: each verified stretch becomes a row carrying the script's own words, and the rest of the recording stays untranscribed rows, so no audio is lost or repeated. Script provenance is in the first word's transcript_source object in words; word times are milliseconds from the row's start. Alignment scores are not ASR confidence. Unmatched scripts/PDFs remain preserved separately; no generated ASR text is passed off as website text. Language is unlabelled until verified; no language is guessed from a station name.

# Use after loading or concatenating configs, before supervised ASR training:
paired = ds.filter(lambda r: r["source_collection"] != "untranscribed")

Every row of every open dataset is included, evaluation sets too: those rows carry their split in source_collection (valid, validation, dev, test, eval, benchmark), so they can be left out of training in one filter.

from datasets import load_dataset

ds = load_dataset("kapturecx/bolAIndia", "yodas3", split="train")
held_out = ('valid', 'validation', 'dev', 'test', 'eval', 'benchmark')
ds = ds.filter(lambda r: r["source_collection"] not in held_out)
ds = ds.filter(lambda r: r["language_code"] == "ta")                     # one language

In open-dataset rows the call-centre columns carry the dataset's own provenance:

column	open-dataset rows
provider	who wrote the transcript: human, captions, asr, read prompts ...
model	the dataset's name
source_db	the original dataset: HF repo id or URL
source_collection	its split: train, or one of the held-out names above
config_id	the dataset's own subset (language, district, config)
mongo_id / conversation_id / call_sid	its utterance, recording and speaker ids
recording_url	where the audio sits inside the original dataset

confidence holds the dataset's own quality score where it has one (NPTEL: text-to-ASR alignment; WorldSpeech: 1 - CER) and is empty otherwise; VAD and phone-hash columns are empty.

Languages

Audio per language across every source, as the transcription system or the dataset labelled it (the language_code column). unlabelled rows carry no language code.

language	code	hours	chunks	share of hours
Hindi	hi	72.5K	39M	48.4%
unlabelled	—	12K	1.5M	8.0%
Bengali	bn	9.9K	4.2M	6.6%
Punjabi	pa	8.6K	2.6M	5.7%
English	en	8.3K	7.3M	5.5%
Tamil	ta	7.2K	2.9M	4.8%
Telugu	te	6K	2.1M	4.0%
Marathi	mr	4.7K	1.9M	3.1%
Malayalam	ml	3.7K	1.7M	2.5%
Kannada	kn	3.7K	1.6M	2.4%
Gujarati	gu	2.1K	997K	1.4%
Odia	or	1.4K	777K	1.0%
Assamese	as	1.3K	706K	0.9%
Urdu	ur	1.3K	712K	0.8%
Nepali	ne	1.1K	492K	0.7%
Maithili	mai	693	361K	0.5%
Bodo	brx	599	354K	0.4%
Sanskrit	sa	591	290K	0.4%
Manipuri	mni	576	301K	0.4%
Dogri	doi	559	240K	0.4%
Sindhi	sd	493	256K	0.3%
Santali	sat	469	257K	0.3%
Kashmiri	ks	413	224K	0.3%
Konkani	kok	354	203K	0.2%
Bhojpuri	bho	330	150K	0.2%
Chhattisgarhi	hne	240	99.8K	0.2%
Magahi	mag	128	75.7K	<0.1%
Chakma	ccp	49.3	40.7K	<0.1%
Garo	grt	44.2	39.1K	<0.1%
Rajasthani	raj	29.7	15.9K	<0.1%
Garhwali	gbm	27.4	8.6K	<0.1%
Haryanvi	bgc	26	8K	<0.1%
Bilaspuri	kfs	22.7	7.2K	<0.1%
Kullu Pahari	kfx	21.2	7K	<0.1%
Nagamese	nag	21.1	14.4K	<0.1%
Mandeali	mjl	20.2	6.8K	<0.1%
Gaddi	gbk	20.1	7.7K	<0.1%
Mizo	lus	20.1	11.2K	<0.1%
Outer Seraji	x-outer-seraji	19.6	7K	<0.1%
Kangri	xnr	19.2	7.2K	<0.1%
Bhadrawahi	bhd	18.2	6.8K	<0.1%
Mahasu Pahari	bfz	17.7	6.9K	<0.1%
Angika	anp	13.6	2.6K	<0.1%
Korwa	kfp	11.4	921	<0.1%
Konkani (individual language)	knn	11.3	890	<0.1%
Tulu	tcy	11.2	2.6K	<0.1%
Wancho	nnp	11	10.4K	<0.1%
Nimadi	noe	10	844	<0.1%
Jaunsari	jns	9.5	857	<0.1%
Kui	uki	9.1	859	<0.1%
Kota	kfe	8.7	707	<0.1%
Bhili	bhb	8.6	1.2K	<0.1%
Marwari	mwr	7.8	4.2K	<0.1%
Kinnauri	kfk	7.5	627	<0.1%
Kokborok	trp	7.4	3.3K	<0.1%
Goan Konkani	gom	6.8	894	<0.1%
Kathoriya Tharu	tkt	5.7	530	<0.1%
Braj	bra	4.8	392	<0.1%
Western Punjabi	pnb	4.6	287	<0.1%
Bajjika	vjk	4.4	2K	<0.1%
Doteli	dty	4.3	479	<0.1%
Rana Tharu	thr	3.6	331	<0.1%
Khortha	x-khortha	3.4	2.5K	<0.1%
Chitwania Tharu	the	3.4	296	<0.1%
Sikkimese	sip	3.4	321	<0.1%
Deccan	dcc	2.8	195	<0.1%
Bundeli	bns	2.5	542	<0.1%
Kumaoni	kfy	2.3	1K	<0.1%
Kochila Tharu	thq	2.3	242	<0.1%
Kanauji	bjj	2.1	193	<0.1%
Halbi	hlb	2	979	<0.1%
Sumi	nsm	1.5	914	<0.1%
Idu Mishmi	clk	1	1.1K	<0.1%
Malvani	x-malvani	0.93	605	<0.1%
Sadri	sck	0.9	660	<0.1%
Maria	mrr	0.82	56	<0.1%
Hadothi	hoj	0.74	46	<0.1%
Mewari	mtr	0.69	58	<0.1%
Od	odk	0.64	50	<0.1%
Surgujia	sgj	0.63	525	<0.1%
Karbi	mjw	0.62	596	<0.1%
Rengma	nre	0.47	352	<0.1%
Ao	njo	0.47	281	<0.1%
Gondi	gon	0.41	268	<0.1%
Northern Hindko	hno	0.4	28	<0.1%
Chakhesang	x-chakhesang	0.35	269	<0.1%
Pahari-Potwari	phr	0.35	23	<0.1%
Wadiyara Koli	kxp	0.29	21	<0.1%
Rongmei	nbu	0.27	190	<0.1%
Awadhi	awa	0.25	172	<0.1%
Kachi Koli	gjk	0.23	16	<0.1%
Kurukh	kru	0.22	131	<0.1%
Surjapuri	sjp	0.21	177	<0.1%
Sambalpuri	spv	0.2	149	<0.1%
Dhundari (Jaipuri)	dhd	0.17	81	<0.1%
Beary	x-beary	0.16	120	<0.1%
Angami	njm	0.16	104	<0.1%
Tagin	tgj	0.12	93	<0.1%
Nyishi	njz	0.11	77	<0.1%
Thethi	x-thethi	0.11	80	<0.1%

NewsOnAIR original archive (separate from training rows)

Full original audio, original scripts/documents, and catalogue snapshots are preserved under archives/newsonair/. Collection is incremental, not complete. These are not aligned audio/transcript pairs and are excluded from all Parquet hours, chunks, language tables, and dataset configs above.

Any standardized newsonair chunks are counted separately in the Parquet tables. They derive from these originals: do not add archive hours to Parquet hours.

Verified archive checkpoint: 2026-10-05T04:49:38.163674+00:00. Statistics refresh periodically; newer archive commits may not yet be reflected in this snapshot.

preserved originals	count
unique audio hours	12,213.53
unique audio files	80,784
unique script/document files	25,990
unique HTML script pages	3,873
unique catalogue snapshots	10,636
all unique objects	121,283
object bytes (excluding manifests)	364,744,671,267
currently published asset URLs (not unique files)	110,807
verified publication manifests	9,081

Objects and audio duration are counted once per SHA-256, including retained versions of changed source URLs. URL aliases do not add another stored copy. Different encodings or overlapping recordings are not deduplicated by this measure. Scripts become transcripts only where their text verifies against the audio (see newsonair above); original rights are not asserted to be an open licence.

Browse originals · Machine-readable archive statistics

Loading

Rows are keyed by chunk_id. In the call-centre configs a chunk that two vendors transcribed differently has one row per vendor (same chunk_id, different provider and text). To keep one transcript per clip, keep the first row of each chunk_id:

seen = set()
ds = ds.filter(lambda r: not (r["chunk_id"] in seen or seen.add(r["chunk_id"])))

from datasets import load_dataset

ds = load_dataset("kapturecx/bolAIndia", "combined", split="train")
print(ds[0]["text"], ds[0]["confidence"])

Fields

field	meaning
audio	16 kHz mono chunk of the human channel
text	provider transcript; empty means nothing intelligible was heard
confidence	provider confidence for the chunk (0-1)
word_confidence_mean / word_confidence_min	aggregated per-word confidence
words	JSON per-word timings and confidences
language_code / language_confidence	detected language and its confidence
is_unintelligible	true when the provider returned no text
provider / model / source_id	which system produced the transcript
speech_ratio / vad_detail	fused and per-engine VAD speech ratios
channel	side of the dual-channel recording (human)
source_db / mongo_id / conversation_id	provenance of the call; resolving these needs access to the source database
recording_url / s3_key / bucket	blank by design — the source recording is not distributed with this dataset
chunk_start_ms / chunk_end_ms	position of the chunk inside that recording

Phone numbers are stored only as salted hashes.