ELSA Realtime Streaming Speech API

openapi: 3.0.1
info:
  title: ELSA Speech Assessment API
  description: >-
    The ELSA API scores recorded or streamed English audio for non-native
    speakers across pronunciation, fluency, intonation, grammar, and vocabulary,
    returning sentence, word, and phoneme-level feedback. ELSA splits the API
    into a scripted variant (v2 — audio plus an expected script) and an
    unscripted variant (v1 — open-ended speech, no script). Each variant offers
    a single-file batch HTTP endpoint and a realtime WebSocket streaming
    endpoint. Access is partner/B2B and gated behind a requested API token;
    only the batch HTTP endpoints are modeled below. The realtime scoring
    surface is delivered over WebSocket (wss://api.elsanow.io/api/v2/ws/score_audio
    for scripted and wss://api.elsanow.io/api/v1/ws/score_audio_plus for
    unscripted) and is documented at the ELSA developer site but not modeled as
    HTTP operations here. Request and response field schemas are partner-gated
    and are not fully published; bodies are described generically and not
    fabricated.
    Disambiguation: this is ELSA Speak / ELSA Corp (elsaspeak.com, elsanow.io),
    not the open-source Elsa .NET workflows library.
  termsOfService: https://www.elsaspeak.com/en/terms-of-service/
  contact:
    name: ELSA API
    url: https://api-external-doc.elsanow.co/
  version: '1.0'
servers:
  - url: https://api.elsanow.io
    description: ELSA batch speech-scoring API host (HTTP). Realtime scoring uses the wss:// scheme on the same host.
security:
  - ElsaAuth: []
paths:
  /api/v2/score_audio:
    post:
      operationId: scoreAudioScripted
      tags:
        - Scripted
      summary: Score a recorded audio file against an expected script (batch).
      description: >-
        Single-file (batch) scripted scoring. Submit a recorded audio file plus
        the expected script/sentence and an API plan; returns sentence, word, and
        phoneme-level pronunciation scores along with mispronunciation hints,
        word stress, reading speed, pausing, fluency, and intonation. Exact
        request and response schemas are partner-gated; see the ELSA developer
        documentation for the authoritative field definitions.
      requestBody:
        required: true
        content:
          multipart/form-data:
            schema:
              type: object
              description: >-
                Multipart form containing at least an api_plan field and an
                audio_file field, plus the expected script/sentence. Full field
                set is documented in the partner API reference.
      responses:
        '200':
          description: Scoring result (scripted). Schema is partner-gated; not modeled here.
  /api/v1/score_audio_plus:
    post:
      operationId: scoreAudioUnscripted
      tags:
        - Unscripted
      summary: Score a recorded audio file with no reference script (batch).
      description: >-
        Single-file (batch) unscripted scoring. Submit spontaneous English audio
        with no reference script and an API plan; returns a spoken transcript plus
        pronunciation, prosody/intonation, fluency, grammar, and vocabulary
        analysis with improvement suggestions. Exact request and response schemas
        are partner-gated; see the ELSA developer documentation for the
        authoritative field definitions.
      requestBody:
        required: true
        content:
          multipart/form-data:
            schema:
              type: object
              description: >-
                Multipart form containing at least an api_plan field and an
                audio_file field. Full field set is documented in the partner
                API reference.
      responses:
        '200':
          description: Scoring result (unscripted). Schema is partner-gated; not modeled here.
components:
  securitySchemes:
    ElsaAuth:
      type: apiKey
      in: header
      name: Authorization
      description: >-
        API token passed in the Authorization header using the ELSA scheme,
        e.g. "Authorization: ELSA <API_token>". Tokens are issued to partners;
        an optional master key can generate tokens programmatically. Request
        access via the ELSA website.