#When using STT Scribe v2 Realtime through the python sdk with mono channel audio, can we configure

1 messages · Page 1 of 1 (latest)

random windBOT
#
AI-generated Answer for epoxyra

Short answer: You can either force one language or let it auto‑detect; there’s no option to pass a list of 2–3 languages.

  • To force a single language, set language_code (ISO‑639‑1/3) when you open the Realtime Scribe v2 WebSocket (the SDK uses the same parameter). (elevenlabs.io)
  • If you omit language_code, Scribe v2 Realtime auto‑detects and can switch languages mid‑conversation. (elevenlabs.io)
  • Realtime Scribe currently expects mono audio; the cookbook notes “Only mono audio is supported at this time.” (elevenlabs.io)

References: Realtime STT API reference and cookbook (query parameters include a single language_code), plus the Scribe v2 Realtime announcement (automatic language detection). (elevenlabs.io)

Sources:
Realtime | ElevenLabs Documentation
Introducing Scribe v2 Realtime
Realtime Speech to Text | ElevenLabs Documentation
Realtime | ElevenLabs Documentation

minor bough
#

Does Scribe v2 realtime stores its data on US servers only or can it store them on a European server ? (like for elevenlabs TTS)

random windBOT
#
AI-generated Answer for epoxyra

Short answer: Yes. By default, ElevenLabs stores customer data in the U.S., but Enterprise customers can opt into EU data residency via an isolated environment. For EU storage, use the EU environment and its endpoints with an EU-issued API key (API: https://api.eu.residency.elevenlabs.io, WebSocket: wss://api.eu.residency.elevenlabs.io). (elevenlabs.io)

Notes specific to Scribe v2 Realtime:

  • Scribe v2 Realtime is “enterprise ready” and explicitly supports EU (and India) data residency options. (elevenlabs.io)
  • Storage will be in the chosen residency location; processing can occur outside that region unless you configure it. For EU, you can keep processing within the EU by using the API with Zero Retention Mode and avoiding out‑of‑region integrations. (elevenlabs.io)

In short: Scribe v2 Realtime data does not have to be stored only in the U.S.—EU storage is available through the Enterprise EU isolated environment. (elevenlabs.io)

Sources:
Data residency | ElevenLabs Documentation
Introducing Scribe v2 Realtime
Data residency | ElevenLabs Documentation
Data residency | ElevenLabs Documentation

minor bough
#

When using STT Scribe v2 Realtime through the python sdk with automatic language detection, how can I get the language that the model detected ?

random windBOT
#
AI-generated Answer for epoxyra

Short answer: it isn’t documented.

  • The Realtime Scribe API supports automatic language detection if you omit language_code when connecting, but the public Realtime event schemas (session_started, partial_transcript, committed_transcript, committed_transcript_with_timestamps) do not document a language field being returned in the stream. (elevenlabs.io)

  • The only place where the detected language is explicitly returned today is on transcript objects retrieved via the REST “Get transcript” endpoint, which includes language_code and language_probability. Example (from the docs):

    • GET /v1/speech-to-text/transcripts/:transcription_id → returns { language_code, language_probability, text, words, … }. (elevenlabs.io)

What you can do now

  • If you must know the language during realtime, there’s no documented way to read it from the streaming events via the Python SDK. Set languageCode when connecting to pin the language instead of relying on auto-detect. (elevenlabs.io)
  • If you can read it after the fact, fetch the transcript via the REST endpoint above and read language_code/language_probability. (elevenlabs.io)

Background: Scribe v2 Realtime advertises automatic language detection and mid-conversation switching, but implementation details for exposing the detected language in realtime aren’t provided in the public docs. (elevenlabs.io)

Sources:
Realtime Speech to Text | ElevenLabs Documentation
Get transcript | ElevenLabs Documentation
Realtime Speech to Text | ElevenLabs Documentation
Get transcript | ElevenLabs Documentation
Introducing Scribe v2 Realtime

minor bough
#

When using STT Scribe v2 Realtime through the python sdk, how come that I often get partial transcripts right after a commited transcript with both having the same content ?