Sarvam AI Speech to Text API

openapi: 3.0.1
info:
  title: Sarvam AI Chat Speech to Text API
  description: REST API for Sarvam AI, India's full-stack sovereign AI platform. Provides chat completions over Sarvam's Indic LLMs, speech-to-text and speech-to-text-translate, text-to-speech, text translation, transliteration, and language identification across Indian languages. All requests authenticate with an api-subscription-key header.
  termsOfService: https://www.sarvam.ai/terms-of-service
  contact:
    name: Sarvam AI Support
    url: https://www.sarvam.ai
  version: '1.0'
servers:
- url: https://api.sarvam.ai
security:
- ApiSubscriptionKey: []
tags:
- name: Speech to Text
paths:
  /speech-to-text:
    post:
      operationId: transcribeSpeech
      tags:
      - Speech to Text
      summary: Transcribe audio to text
      description: Transcribes an audio file in an Indian language using the Saaras or Saarika speech-recognition models. Supports auto language detection, word-level timestamps, and speaker diarization.
      requestBody:
        required: true
        content:
          multipart/form-data:
            schema:
              $ref: '#/components/schemas/SpeechToTextRequest'
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/SpeechToTextResponse'
        '401':
          $ref: '#/components/responses/Unauthorized'
        '429':
          $ref: '#/components/responses/TooManyRequests'
  /speech-to-text-translate:
    post:
      operationId: translateSpeech
      tags:
      - Speech to Text
      summary: Transcribe and translate audio to English
      description: Transcribes Indic-language audio and translates the result into English in a single request using the Saaras family of models.
      requestBody:
        required: true
        content:
          multipart/form-data:
            schema:
              $ref: '#/components/schemas/SpeechToTextTranslateRequest'
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/SpeechToTextResponse'
        '401':
          $ref: '#/components/responses/Unauthorized'
        '429':
          $ref: '#/components/responses/TooManyRequests'
components:
  schemas:
    SpeechToTextTranslateRequest:
      type: object
      required:
      - file
      properties:
        file:
          type: string
          format: binary
          description: Audio file to transcribe and translate to English.
        model:
          type: string
          enum:
          - saaras:v2.5
          - saaras:v3
          default: saaras:v2.5
        prompt:
          type: string
          description: Optional context to improve translation quality.
    Error:
      type: object
      properties:
        error:
          type: object
          properties:
            message:
              type: string
            code:
              type: string
    SpeechToTextRequest:
      type: object
      required:
      - file
      properties:
        file:
          type: string
          format: binary
          description: Audio file (WAV, MP3, AAC, AIFF, OGG, OPUS, FLAC, MP4/M4A, AMR, WMA, WebM, PCM). Optimized for 16kHz.
        model:
          type: string
          enum:
          - saarika:v2.5
          - saaras:v3
          default: saarika:v2.5
        language_code:
          type: string
          description: BCP-47 code (e.g. hi-IN) or "unknown" for auto-detection.
        mode:
          type: string
          description: saaras:v3 only.
          enum:
          - transcribe
          - translate
          - verbatim
          - translit
          - codemix
        input_audio_codec:
          type: string
          description: Required only for PCM files.
          enum:
          - pcm_s16le
          - pcm_l16
          - pcm_raw
    SpeechToTextResponse:
      type: object
      properties:
        request_id:
          type: string
        transcript:
          type: string
        language_code:
          type: string
        language_probability:
          type: number
        timestamps:
          type: object
          description: Word-level timing data when requested.
        diarized_transcript:
          type: object
          description: Speaker-attributed segments when diarization is enabled.
  responses:
    Unauthorized:
      description: Missing or invalid api-subscription-key.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
    TooManyRequests:
      description: Rate limit exceeded.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
  securitySchemes:
    ApiSubscriptionKey:
      type: apiKey
      in: header
      name: api-subscription-key
      description: API subscription key created in the Sarvam AI dashboard.