Skevyr

OTS Data

Ready-to-use datasets for AI in Indian languages

Off-the-shelf (OTS) speech, text and image datasets from Skevyr: collected with consent, checked by people, and cleared for commercial use. Faster and cheaper than a custom collection.

Request a sample or quote

Available datasets

Volumes, speaker counts and pricing are shared on request with a free sample.

AvailableSpeech · Hindi

Hindi read speech

Native Hindi speakers reading everyday sentences, recorded on phones and laptops in real homes.

  • Balanced across gender and age groups
  • Speakers from several states
  • WAV / FLAC, 16 kHz, with transcripts
  • Every recording manually reviewed
Request a sample →
AvailableSpeech · Bengali

Bengali conversational speech

Natural two-person conversations in Bengali on everyday topics.

  • Speaker-separated channels
  • Time-stamped transcripts
  • West Bengal and nearby regions
Request a sample →
AvailableSpeech + transcripts · Hindi + English

Hindi-English (Hinglish) transcribed speech

Code-mixed speech as people really talk, with verbatim transcripts in Devanagari and Latin script.

  • Verbatim and normalised transcripts
  • Noise and speaker tags
  • Customer-care and daily-life topics
Request a sample →
AvailableSpeech · Tamil

Tamil read speech

Native Tamil speakers reading prompts across varied recording conditions.

  • Tamil Nadu and Puducherry speakers
  • WAV / FLAC, 16 kHz, with transcripts
Request a sample →
AvailableSpeech · Marathi

Marathi voice commands

Short commands and wake-word style phrases for voice assistants and IVR.

  • Multiple rooms and distances per speaker
  • Near-field and far-field
Request a sample →
AvailableImages · Any

Indian retail shelf images

Photos of grocery and kirana store shelves across Indian cities, ready for product detection.

  • Bounding-box annotation available
  • Varied lighting and store formats
  • No people's faces
Request a sample →

Coming soon

Coming soonSpeech · Telugu

Telugu read speech

Native Telugu speakers reading everyday sentences.

  • Collection in progress
Register interest →

How licensing works

  1. 1Tell us what you needLanguage, data type, volume and use case.
  2. 2Get a free sampleA small sample with the full metadata, so you can check quality and format.
  3. 3Agree the licenceCommercial licence with clear terms. All data is collected with consent.
  4. 4Receive the dataSecure delivery in your format: WAV/FLAC, JSON/CSV manifests, transcripts.

Request a sample or quote

Tell us what you're building. Our data team replies within 1-2 working days.

Languages, hours or volume, use case, and whether you need a sample first.