Skip to main content
Vyasa is SomyaLabs’ automatic speech recognition (ASR) model, built for Indian languages, accents, and real-world audio. Transcribe audio with POST /v1/speech/transcriptions.

Request

Send the audio as multipart/form-data under the audio field:

Response

The standard envelope includes the transcript and model-predicted language:
predicted_lang is the detected language ID. It can be null if the model server does not provide it. Unlimited test API keys return credits_used: 0.0 and credits_remaining: null.

Audio formats & limits

The endpoint accepts WAV, MP3, OGG, and WebM audio, up to 25 MB. WebM is converted to WAV before inference. Oversized or unsupported uploads are rejected:
  • 413 — the audio file is too large.
  • 422 — the audio format isn’t supported or the audio is invalid.
If your file is rejected, convert it to WAV and retry. See Errors for handling non-200 responses.