Saaras | Sarvam API Docs

Saaras is our state-of-the-art speech recognition family with flexible output formats. It supports multiple output modes including transcription, translation, verbatim, transliteration, and code-mixed outputs. Saaras is built to make Indic languages LLM-comprehensible, offering accurate transcriptions and translations across 23 languages (22 Indian languages + English).

Saaras v4 is the latest model, adding Global English support (in addition to Indian English) across all output modes while keeping the same 22 Indic language coverage. Saaras v3 remains the default, recommended model. Both are available on the Speech-to-Text endpoint (/speech-to-text) and support multiple output modes via the mode parameter.

At a Glance


Model ID	`saaras:v3` (default, recommended), `saaras:v4` (latest)
What it does	Speech-to-text with five output modes: transcribe, translate, verbatim, translit, codemix
Languages	23 (22 Indian + English), automatic language detection (full list). `saaras:v4` additionally supports Global English alongside Indian English
APIs	REST (≤30 s), Batch (≤2 hr/file), WebSocket streaming
Input limits	30 s per REST request; WAV, MP3, AAC, FLAC, OGG and more, all limits
Pricing	Pricing page
Best for	Voice agents, call analytics, 8 kHz telephony audio, code-mixed speech
Known limitations	See below

Saaras v3 vs v4

	Saaras v3	Saaras v4
Status	Default, recommended	Latest
Output modes	transcribe, translate, verbatim, translit, codemix	transcribe, translate, verbatim, translit, codemix
English support	Indian English (`en-IN`)	Indian English and Global English
Endpoint	`/speech-to-text`	`/speech-to-text`

To try the latest model, set model="saaras:v4" in any of the requests below. The request/response shape is identical to saaras:v3.

saaras:v4 also supports Keyterm Prompting: a list of up to 50 domain-specific names, places, brands, or technical terms to bias recognition toward. The examples below pass keyterms — it is accepted only on saaras:v4, on the REST and Batch APIs, so drop that argument when using saaras:v3 or the streaming endpoint. See Keyterm Prompting.

Output Modes

Saaras supports multiple output modes via the mode parameter, available on both saaras:v3 and saaras:v4. Each mode produces different output formats for the same input audio.

Example audio: “मेरा फोन नंबर है 9840950950”

Mode	Description	Example Output
`transcribe` (default)	Standard transcription in the original language with proper formatting and number normalization	`मेरा फोन नंबर है 9840950950`
`translate`	Translates speech from any supported Indic language to English	`My phone number is 9840950950`
`verbatim`	Exact word-for-word transcription without normalization, preserving filler words and spoken numbers as-is	`मेरा फोन नंबर है नौ आठ चार zero नौ पांच zero नौ पांच zero`
`translit`	Romanization - Transliterates speech to Latin/Roman script	`mera phone number hai 9840950950`
`codemix`	Code-mixed text with English words in English and Indic words in native script	`मेरा phone number है 9840950950`

Key Features

Domain-Aware Translation

Advanced prompting system for domain-specific translation and hotword retention, ensuring accurate context preservation.

Superior Telephony Performance

Optimized for 8KHz telephony audio with enhanced multi-speaker recognition capabilities.

Intelligent Entity Preservation

Preserves proper nouns and entities accurately across languages, maintaining context and meaning.

Multi-Language Support

Supports 23 languages (22 Indian + English) with optional language identification.

Speaker Diarization

Provides diarized outputs with precise timestamps for multi-speaker conversations through batch API.

Direct Translation

Converts speech directly to English text, eliminating the need for separate transcription and translation steps.

Global English (v4)

saaras:v4 extends English recognition beyond Indian English (en-IN) to Global English accents, on top of the same 22 Indic languages.

Language Support

Saaras v3 supports 23 languages (22 Indian languages + English) with comprehensive dialect and accent coverage, including code-mixed audio support and intelligent proper noun preservation for speech-to-English translation.

Language	Language Code	Language	Language Code
Hindi	`hi-IN`	Assamese	`as-IN`
Bengali	`bn-IN`	Urdu	`ur-IN`
Kannada	`kn-IN`	Nepali	`ne-IN`
Malayalam	`ml-IN`	Konkani	`kok-IN`
Marathi	`mr-IN`	Kashmiri	`ks-IN`
Odia	`od-IN`	Sindhi	`sd-IN`
Punjabi	`pa-IN`	Sanskrit	`sa-IN`
Tamil	`ta-IN`	Santali	`sat-IN`
Telugu	`te-IN`	Manipuri	`mni-IN`
English	`en-IN`	Bodo	`brx-IN`
Gujarati	`gu-IN`	Maithili	`mai-IN`
		Dogri	`doi-IN`

Language codes are optional. When not specified or set to unknown, the model will automatically detect the input language and return a language_probability score indicating detection confidence.

Additional Capabilities:

Includes dialects and accents of the above languages
Code-mixed audio support
Intelligent Proper Noun and Entity Preservation to ensure proper nouns, regional names, and entities are recognized and retained accurately during transcription

API Response Format

The Speech-to-Text API returns a JSON response with the following fields:

Field	Type	Description
`request_id`	`string`	Unique identifier for the API request.
`transcript`	`string`	The transcribed text from the provided audio file.
`timestamps`	`object or null`	Contains chunk-level timestamps (`start_time_seconds`, `end_time_seconds`, `words`, each entry is a sentence/phrase, not an individual word). Only included when `with_timestamps` is set to `true`. Word-level timestamps are not supported.
`diarized_transcript`	`object or null`	Diarized transcript with speaker labels. Available through batch API.
`language_code`	`string or null`	BCP-47 code of the detected language (e.g., `hi-IN`). Returns the most predominant language if multiple are detected. Returns `null` if no language is detected.
`language_probability`	`number or null`	Float value (0.0 to 1.0) indicating the probability of the detected language being correct. Higher values indicate higher confidence. Returns a value when `language_code` is not provided or set to `unknown`. Returns `null` when a specific `language_code` is provided (language detection is skipped). Always present in the response.

Example Response:

{
  "request_id": "20260209_abc123-def4-5678-ghij-klmnopqrstuv",
  "transcript": "नमस्ते, आप कैसे हैं?",
  "timestamps": null,
  "diarized_transcript": null,
  "language_code": "hi-IN",
  "language_probability": 0.95
}

Key Capabilities

Transcribe Mode

Translate Mode

Verbatim Mode

Translit Mode

Codemix Mode

Standard transcription in the original language with proper formatting and number normalization. This is the default mode.

from sarvamai import SarvamAI
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",  # default mode
        keyterms=["Sarvam", "New Delhi", "Vistaar"]
    )
    print(response.transcript)
    # Output: मेरा फोन नंबर है 9840950950
except Exception as e:
    print(f"Error: {e}")

Limits

Limit	Value
Max audio duration (real-time REST)	30 seconds
Supported formats	WAV, MP3, AAC, AIFF, OGG, OPUS, FLAC, MP4, AMR, WMA, WebM (auto-detected)
Raw PCM input (`pcm_s16le`, `pcm_l16`, `pcm_raw`)	Requires `input_audio_codec`; must be 16 kHz
Longer audio	Use the Batch API (up to 2 hours per file)
Rate limits	See Rate Limits

Known Limitations

Limitation	Detail	Workaround
30-second cap on real-time REST	The real-time `/speech-to-text` endpoint only accepts audio up to 30 seconds long	Use the Batch API for longer recordings (up to 2 hours per file)
`mode` parameter is v3/v4-only	The `mode` parameter (transcribe / translate / verbatim / codemix) works only with `saaras:v3` and `saaras:v4`	Use `saaras:v3` or `saaras:v4`; older versions ignore `mode`
Translate mode outputs English only	`mode="translate"` always produces English text, regardless of input language	Use transcribe mode if you need output in the original language

Next Steps

Developer quickstart

Learn how to integrate Saaras v3 into your application.

API Reference

Complete API documentation for Speech-to-Text endpoint.

Cookbook

Step-by-step tutorial for speech-to-text transcription.

Saaras v2.5 (Deprecated Soon)

Deprecation Notice: Saaras v2.5 will be deprecated soon. We recommend migrating to Saaras v3 for improved accuracy and performance. The v2.5 model will continue to work during the transition period, but new features and improvements will only be available in v3.

About Saaras v2.5

Saaras v2.5 is the previous speech translation model available in the Speech-to-Text Translate endpoint (/speech-to-text-translate). It converts speech directly to English text with enhanced telephony support and intelligent entity preservation.

Key Difference: Saaras v2.5 uses the /speech-to-text-translate endpoint, while Saaras v3 uses the /speech-to-text endpoint with mode parameter support.

Key Features (v2.5)

Domain-Aware Translation

Advanced prompting system for domain-specific translation and hotword retention, ensuring accurate context preservation.

Superior Telephony Performance

Optimized for 8KHz telephony audio with enhanced multi-speaker recognition capabilities.

Intelligent Entity Preservation

Preserves proper nouns and entities accurately across languages, maintaining context and meaning.

Multi-Language Support

Supports 11 Indian languages with optional language identification.

Speaker Diarization

Provides diarized outputs with precise timestamps for multi-speaker conversations through batch API.

Direct Translation

Converts speech directly to English text, eliminating the need for separate transcription and translation steps.

Translation Quality (v2.5 Benchmarks)

COMET score, a robust metric for evaluating machine speech-translations, assesses semantic accuracy, fluency, and contextual relevance. Saaras v2.5 achieves exceptional performance on the Vistaar+Indicvoices Benchmark, a dataset curated from diverse Indian language audio sources, including code-mixed content, noisy environments, and regional accents.

COMET Score Performance:

Across 11 Languages: 89.3%
English: 94.62%
Hindi: 91.83%
9 Other languages: 88.41%

Higher is better; Compared on VISTAAR + IndicVoices Benchmark

Why COMET? It evaluates not only lexical accuracy but also how well the translation captures meaning and context, critical for Indic languages with complex structures.

Dataset Description: Contains real-world, multi-accented speech samples that covers 10 major Indic languages, ensuring representation of India’s linguistic diversity. Includes code-mixed phrases, domain-specific vocabulary, and colloquial expressions.

Migration Guide

To migrate from Saaras v2.5 to v3:

Change the endpoint: Switch from /speech-to-text-translate to /speech-to-text
Update the model parameter: Change from saaras:v2.5 to saaras:v3
Add the mode parameter: Use mode="translate" to get English output (similar to v2.5 behavior)

# Endpoint change
- POST /speech-to-text-translate
+ POST /speech-to-text
# Parameter changes
- model="saaras:v2.5"
+ model="saaras:v4"
+ mode="translate"

SDK Migration:

# Python
- response = client.speech_to_text.translate(
-     file=open("audio.wav", "rb"), model="saaras:v2.5"
- )
+ response = client.speech_to_text.transcribe(
+     file=open("audio.wav", "rb"), model="saaras:v4", mode="translate"
+ )
# JavaScript
- const response = await client.speechToText.translate({ file: fs.createReadStream("audio.wav"), model: "saaras:v2.5" });
+ const response = await client.speechToText.transcribe({ file: fs.createReadStream("audio.wav"), model: "saaras:v4", mode: "translate" });

The response format remains compatible.