Speech-to-Text Rest API | Sarvam API Docs

Synchronous Processing

Process short audio files with immediate response. Best for quick transcriptions and testing with a maximum duration of 30 seconds.

Saaras: State-of-the-Art Speech Recognition (Recommended)

Saaras is our state-of-the-art speech recognition family with flexible output formats. It supports multiple modes for different use cases: transcribe, translate, verbatim, transliterate, and codemix.

Recommended for new integrations. Saaras v4 is the default, recommended model, adding Global English support and keyterm prompting. Saaras v3 remains available. Both accept the same request shape, so switching is a one-line change. Learn more about Saaras.

Output Modes

Mode	Description
`transcribe` (default)	Standard transcription in the original language
`translate`	Translates speech to English
`verbatim`	Exact word-for-word transcription
`translit`	Romanization to Latin script
`codemix`	Code-mixed text output

Code Examples

Saaras v3

Saaras v4

Still available. Supports all five output modes across 22 Indian languages plus Indian English.

from sarvamai import SarvamAI
client = SarvamAI(
    api_subscription_key="YOUR_SARVAM_API_KEY",
)
# Transcribe mode (default)
response = client.speech_to_text.transcribe(
    file=open("audio.wav", "rb"),
    model="saaras:v3",
    mode="transcribe"  # or "translate", "verbatim", "translit", "codemix"
)
print(response)

Check out our detailed API Reference to explore all available options.

Preparing Your Audio

Most failed STT requests are caused by the audio itself, not the API call. Run through this checklist before uploading:

Check	Recommendation
Duration	The sync REST API accepts up to 30 seconds per request. For longer files, use the Batch API (up to 2 hours per file) or split the audio into ≤30s chunks.
Sample rate	16 kHz is recommended. 8 kHz telephony audio (IVR, call recordings) is fully supported, no need to upsample.
Channels	Use mono. For stereo telephony recordings with one speaker per channel, split the channels and transcribe each separately to keep speakers separated.
Format	WAV, MP3, AAC, FLAC, or OGG. Prefer WAV (16-bit PCM) for best accuracy.
File integrity	Verify the file exists and is non-empty before uploading (`file.size > 0` in browsers). Pass a file object, not a path string, e.g. `file=open("audio.wav", "rb")` in Python.

Not sure whether to use REST, Batch, or Streaming? See the Which API to Use decision table for a side-by-side comparison of limits, latency, and features.

Need Saaras v4 to recognize specific names, places, brands, or technical terms? Pass a keyterms list to bias recognition. See Keyterm Prompting.

Legacy Model (Deprecated Soon)

The following model will be deprecated soon. We recommend migrating to Saaras v3 for new integrations.

Saaras v2.5: Speech to Text Translation

Saaras v2.5 is available in the Speech-to-Text Translate endpoint for translating speech directly to English.

Deprecation Notice: Saaras v2.5 will be deprecated soon. Use Saaras v3 with mode="translate" instead. See the code examples above.

API Response Format

Speech to Text Transcription Response

Field	Type	Description
`request_id`	string	Unique identifier for the request
`transcript`	string	The transcribed text from the audio file
`language_code`	string	BCP-47 language code of detected language (e.g., `hi-IN`). Returns `null` if no language detected

{
  "request_id": "20241115_12345678-1234-5678-1234-567812345678",
  "transcript": "नमस्ते, आप कैसे हैं?",
  "language_code": "hi-IN"
}

Speech to Text Translation Response

Field	Type	Description
`request_id`	string	Unique identifier for the request
`transcript`	string	Translated text in English
`language_code`	string	BCP-47 code of the detected source language

Supported source languages: hi-IN, bn-IN, kn-IN, ml-IN, mr-IN, od-IN, pa-IN, ta-IN, te-IN, gu-IN, en-IN

{
  "request_id": "20241115_12345678-1234-5678-1234-567812345678",
  "transcript": "Hello, how are you?",
  "language_code": "hi-IN"
}

Error Responses

All errors return a JSON object with an error field (message, code, request_id). The full error-code table, retry guidance, and SDK exception reference live on the central Errors & Troubleshooting page.

Errors specific to this endpoint:

HTTP Status	Error Code	When This Happens	What To Do
`422`	`unprocessable_entity_error`	Invalid audio format, file too large, or audio over 30 seconds	Use supported formats (WAV, MP3, AAC, FLAC, OGG); for longer audio use the Batch API

Error Handling Code Example

from sarvamai import SarvamAI
from sarvamai.core.api_error import ApiError
client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY")
try:
    response = client.speech_to_text.transcribe(
        file=open("audio.wav", "rb"),
        model="saaras:v4",
        mode="transcribe"
    )
    print(response.transcript)
except ApiError as e:
    if e.status_code == 400:
        print(f"Bad request: {e.body}")
    elif e.status_code == 403:
        print("Invalid API key. Check your credentials.")
    elif e.status_code == 429:
        print("Rate limit exceeded. Wait and retry.")
    elif e.status_code == 503:
        print("Service overloaded. Retry with backoff.")
    else:
        print(f"Error {e.status_code}: {e.body}")

Next Steps

Get API Key

Test Integration

Try the API with sample audio files.

Go Live

Deploy your integration and monitor usage.

Need help? Contact us on discord for guidance.