generated: '2026-07-19'
method: searched
source: https://sdk-docs.krisp.ai/docs/voice-translation-api.md
docs: https://sdk-docs.krisp.ai/
authentication:
  style: api-key
  header: 'Authorization: api-key {API_KEY}'
  streaming: session key on the WSS query string
  detail: authentication/krisp-authentication.yml
idempotency:
  supported: false
  note: >-
    Krisp publishes no idempotency key, no Idempotency-Key header, and no retry-safety contract. The
    developer surface is a streaming session API plus two read-only GET operations, so there are no
    non-idempotent write operations that would require one. No Idempotency pointer is emitted.
pagination:
  supported: false
  note: >-
    No paginated collection endpoints are documented. The list-languages operation returns the full list
    in a single response and the docs explicitly advise fetching it at session start rather than caching.
versioning:
  scheme: uri-path
  current: v2
  example: https://api.developers.krisp.ai/v2/sdk/versions/{id}/download-urls
  note: >-
    The list-languages endpoint is documented in the API reference without a version prefix
    (GET /voice-translation/languages), so version prefixing is not applied uniformly across the
    published surface.
  sdk_versioning: semver per SDK family and language binding, tracked in the changelog
  changelog: https://sdk-docs.krisp.ai/changelog
request_tracing:
  client_supplied: metadata.reference_id
  client_supplied_note: >-
    An optional correlation id supplied in the session config metadata and echoed back on every
    transcript, translation, and error event.
  server_supplied: req_id
  server_supplied_note: Returned on REST responses; quote it to support.
  utterance_correlation: utterance_id
  utterance_note: Links interim transcript events to their final transcript and translation events.
error_envelope:
  format: custom
  rfc9457: false
  shape: '{"error": {"code", "reason", "description", "reference_id"}}'
  detail: errors/krisp-error-codes.yml
rate_limit_signaling:
  headers_published: false
  signal: 429 Too Many Requests error event (rate limit exceeded or max concurrent connections exceeded)
  numeric_limits_published: false
content_negotiation:
  rest: application/json
  streaming:
    control: JSON text frames (WebSocket opcode 0x1)
    media: binary frames (WebSocket opcode 0x2)
    audio_formats: [pcm_s16le, opus]
    sample_rate: 16000
    channels: 1
    frame_duration_ms: 20
    note: >-
      Clients must pace audio in real time — one PCM chunk per input_frame_duration ms. At 16 kHz mono
      s16le that is 640 bytes per 20 ms chunk.
session_protocol:
  handshake: >-
    Open wss://streaming.krisp.ai/vt?authorization=Api-Key SESSION_KEY, then send exactly one JSON config
    message before any audio frame.
  config_required: [source_language, target_language, voice]
  config_optional: [audio.format, audio.sample_rate, audio.channels, vocabulary, translation_dictionary,
    transcript.interim, transcript.final, transcript.translate, features.background_voice_cancellation,
    metadata.reference_id]
  teardown: >-
    Drain trailing events (allow ~2s for the final transcript, translation, and audio to land) before
    closing. Close is idempotent.
localization:
  language_codes: BCP 47 (RFC 5646)
  note: Use full locale codes such as en-US, not bare en. The supported list is dynamic.
  reference: https://sdk-docs.krisp.ai/docs/supported-languages.md
related:
- authentication/krisp-authentication.yml
- errors/krisp-error-codes.yml
- lifecycle/krisp-lifecycle.yml
- asyncapi/krisp-voice-translation-asyncapi.yml