Speech in.
Text or speech out.

Sign in to Swarit Cloud → Developers, configure your integration and complete the OAuth authorization-code flow with PKCE. Request vn:ai:execute for API resource https://cloud.swarit.ai/api/v1. Keep access and refresh tokens on your backend.

In Cloud Billing, set an additional-usage spending limit. Standalone speech is postpaid and does not consume your realtime plan’s included minutes. Read the authentication guide →

Upload raw, mono, 16-kHz signed PCM16 little-endian audio as audio/pcm, up to five minutes per request. English is the default; set the recording’s language explicitly.

For a WAV or MP3 recording, convert it first:

ffmpeg -i recording.wav -ac 1 -ar 16000 -f s16le recording.pcm

curl --no-buffer 'https://cloud.swarit.ai/api/voice/v1/audio/transcriptions?language=en' \
  -H "Authorization: Bearer $SWARIT_ACCESS_TOKEN" \
  -H 'Content-Type: audio/pcm' \
  --data-binary @recording.pcm

The response is NDJSON: transcript.partial, transcript.final, then transcription.done with audio_milliseconds. Use a backend HTTP client supporting duplex streaming to read transcripts while uploading. A terminal error event means the request did not complete; HTTP 200 alone is not a completion signal.

Send complete text as JSON, up to 10,000 UTF-16 code units. The response streams raw, mono, 24-kHz signed PCM16 little-endian audio. This endpoint accepts complete text per request; streaming text input belongs to the component playground.

curl 'https://cloud.swarit.ai/api/voice/v1/audio/speech' \
  -H "Authorization: Bearer $SWARIT_ACCESS_TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{"text":"Hello. How can I help you?","language":"en","voice":"aditi"}' \
  --output speech.pcm

ffplay -f s16le -ar 24000 -ac 1 speech.pcm

Choose one of the 22 named voices, including aditi, kabir, kavya and kiran. The default is Aditi. female and male remain compatibility aliases for Aditi and Kabir.

Supported language codes: English en, Hindi hi, Marathi mr, Malayalam ml, Kannada kn, Tamil ta, Telugu te, Gujarati gu, Punjabi pa, Bengali bn, Assamese as and Odia or.

STT: ₹0.245 per input audio minute. Completed input duration is metered; failed or cancelled requests without a completed transcription are not settled as successful transcription.

TTS: ₹1 per 1,000 Unicode code points. Accepted text is metered once audio output starts. Prices exclude 18% GST. Charges use the existing postpaid spending-limit and invoice flow; realtime minute plans remain unchanged. Fractional usage is rounded up only to the final paise charge for each request.

Requests wait for shared GPU capacity for up to 60 seconds; queue capacity is bounded. A request can run for up to 420 seconds. Cancel the request to stop its stream and release capacity. HTTP 401/403 concerns authorization, 402 requires a sufficient spending limit, 400/413 concerns input, and 503 means capacity or runtime is unavailable.

OpenAPI specification → · Current API pricing →