Which languages does Sarvam's voice-to-text support?

Saaras supports all 22 scheduled Indian languages plus English, including Hindi, Tamil, Telugu, Bengali, Marathi, Kannada, Malayalam, Gujarati, Punjabi, Urdu, Odia, Assamese, and more. Code-mixed audio between any of these languages and English works out of the box.

How accurate is it on noisy or low-quality audio?

Saaras is trained on real-world Indian audio, including 8kHz call-center recordings, background noise, multiple speakers, and dialectal variation. Accuracy holds up well in conditions where general-purpose global STT systems degrade significantly.

Does it handle code-mixed speech, like Hindi-English?

Yes. Code-mixing isn't a special mode. It's the default. Mid-sentence switches between Indian languages and English are transcribed in their native scripts (or as Romanised text, if you prefer), preserving the speaker's meaning.

What audio formats and durations are supported?

Common formats: WAV, MP3, OGG, FLAC, and most container formats your application emits. The batch API handles long-form audio; real-time streaming chunks audio over a WebSocket as it arrives. See the API docs for current limits.

Is real-time streaming available, or only batch?

Both. Real-time streaming over WebSockets returns partial transcripts within a few hundred milliseconds, suitable for live agents, captions, and voice-controlled apps. Batch REST handles archived audio, podcasts, and long recordings.

Can it run on-prem or inside our VPC?

Yes. Sarvam offers managed cloud, single-tenant VPC deployment, and on-prem installations. Talk to us if you have data-residency, compliance, or air-gapped network requirements. We have customers in each of those situations.

How does Sarvam handle data privacy and residency?

Indian data residency by default on the managed plan. We do not train on customer audio without an explicit opt-in agreement. For VPC and on-prem deployments your audio never leaves your infrastructure.

How quickly can we integrate it?

If you've integrated any cloud STT before, plan on hours, not weeks. The REST API is conventional, the streaming protocol is documented, and SDKs cover the common languages. Most teams have a working prototype the same day they get an API key.

Voice to text

The most accurate Voice to Text Converter

Sarvam converts voice to text across 22 Indian languages, holding accuracy even when speakers switch languages mid-sentence.

Speak to see live captions

Want to use this API?

Voice to text in the real world

Production-ready for real conversations, background noise, and language switching.

हाँमैंनेPROPER NOUNFlipkartसेNUMBER₹12,499काENTITYSamsungGalaxyCODE SWITCHफ़ोनऑर्डरकियाथा।ऑर्डरIDहैABBREVIATIONFLK-78234-XN।CODE SWITCHडिलीवरीएड्रेसहै42,PROPER NOUNKoramangala5thBlock,PROPER NOUNBangalore।मेराफ़ोननंबरहैPHONE NUMBER9840950950।CODE SWITCHप्लीज़डिलीवरीजल्दीकरदीजिए,मेरापुरानाफ़ोनबिल्कुलकामनहींकररहा।

✱01

हाँमैंनेPROPER NOUNFlipkartसेNUMBER₹12,499काENTITYSamsungGalaxyCODE SWITCHफ़ोनऑर्डरकियाथा।