Dataset · hi-Latn-IN · Spontaneous Speech
Buy a Hinglish spontaneous speech dataset
Speakers talk unscripted on prompted topics — describing a process, recounting an event, arguing a position — with a moderator who prompts but does not lead. Available for Hinglish as a licensed build against your specification, with audio, verbatim transcripts, speaker metadata and consent records.
- Language
- Hinglish (Devanagari + Latin)
- Type
- Spontaneous Speech
- Per speaker
- 25–40 minutes of accepted audio per speaker
- Accepted yield
- 60–70% of recorded time is accepted
How Hinglish spontaneous speech is recorded
Speakers talk unscripted on prompted topics — describing a process, recounting an event, arguing a position — with a moderator who prompts but does not lead.
A 60-minute session covering six to eight topics, recorded continuously and segmented afterwards into utterances at natural pause boundaries.
60–70% of recorded time is accepted. Long silences, moderator speech and abandoned topics are cut in post, so recorded hours run well ahead of delivered hours. Budget for the gap.
What is in the package
- 16-bit PCM WAV audio, captured at 48 kHz / 24-bit and delivered at your target rate
- Verbatim Devanagari + Latin transcripts with utterance-level timestamps
- Filled-pause and false-start markers
- Segment offsets into the parent session file
- Topic label per segment
- Speech-rate and pause-density statistics per speaker
- Per-speaker metadata: age band, gender, region, dialect, education band, device
- Consent records mapped to speaker IDs and retained for audit
- Speaker-disjoint train / dev / test splits
Specification
| Field | Standard value |
|---|---|
| Language | Hinglish (hi-Latn-IN) |
| Speech type | Spontaneous Speech |
| Per speaker | 25–40 minutes of accepted audio per speaker |
| File granularity | Session-length WAV plus segmented utterance files with offsets into the parent |
| Segmentation | Pause-boundary segmentation, reviewed by a native listener rather than left to VAD |
| Moderator channel | Recorded separately and excluded from the delivered speaker audio |
| Disfluency convention | Filled pauses, repetitions and false starts transcribed, not normalised away |
| Dialects covered | Delhi corporate Hinglish, Mumbai Bambaiya, Call-centre register, Youth/social media register |
| Recording cities | Delhi, Gurugram, Noida, Mumbai |
| Transcription | Native verbatim, second-pass reviewed |
Quality gates applied to spontaneous speech
The failure modes this style is prone to: speakers who dry up after a minute, leaving sessions that look complete by duration but are mostly silence
- Segment boundary review — bad boundaries produce truncated words that poison training more than they help
- Disfluency transcription consistency audited across transcribers, since conventions drift fast on this style
- Moderator speech confirmed absent from delivered segments
- Topic distribution checked per speaker so one dominant subject does not skew the language model
Pricing and volumes
| Volume | Indicative range | Delivery |
|---|---|---|
| 50 hours (pilot-scale) | ₹1,60,000 – ₹3,25,000 | 3-4 weeks |
| 250 hours | ₹7,60,000 – ₹15,43,750 | 5-8 weeks |
| 1,000 hours | ₹27,20,000 – ₹55,25,000 | 10-16 weeks |
Bands, not a price list. A signed specification converts them into one fixed figure, and staged delivery lets you start training before the full corpus lands.
Why generic Hinglish corpora underperform
Almost no public corpus contains genuine intra-sentential Hindi-English switching with per-token language tags. This is the highest-value gap for anyone building Indian conversational AI.
Hinglish is the code-mixing case itself. Typical urban customer-support speech is 30-60% English tokens embedded in Hindi grammar, with switching several times per utterance.
- Whether English tokens are written in Latin or transliterated into Devanagari must be fixed by rule, not left to annotators
- Language-ID tagging per token is required for training but is skipped by most vendors
- Ambiguous words shared by both languages need an explicit tie-break rule
Licensing and compliance
- Perpetual, transferable licence covering model training, derivative models and commercial deployment
- Written informed consent from every speaker, retained and auditable
- Compliant with India's DPDP Act 2023 handling requirements and mapped to GDPR obligations for EU buyers
- No scraped, resold or re-licensed third-party audio in any delivery
How buying works
- Send your specification or use the standard one above
- Receive a fixed scope, price and delivery schedule within one working day
- Optional paid pilot of 10-20 hours against the same spec
- Recruitment and recording begin on signature, with staged batches
- Acceptance testing per batch, then final delivery with the full manifest
Frequently asked
Can I buy an off-the-shelf Hinglish spontaneous speech dataset today?
We build to specification rather than reselling a fixed corpus, because off-the-shelf Hinglish audio almost never matches a real deployment's dialect mix and noise profile. A 50-hour build ships in three to four weeks.
What licence do we get?
A perpetual, transferable licence with full IP transfer on delivery. You can train, distribute and commercialise models built on the data without further payment.
Can we get a free sample?
Yes — a short sample reflecting the agreed specification is provided before the pilot is signed, so you can check transcript quality and audio characteristics.
Which Hinglish dialects are included?
By default Delhi corporate Hinglish, Mumbai Bambaiya, Call-centre register and others, with quotas set at recruitment. You can weight the mix to match where your product is deployed.
Related pages
Quote this Hinglish dataset
Tell us the volume and dialect mix for Hinglish spontaneous speech and you get a fixed price and delivery date.