how i can control model language in sts setup LiveKit Community #ask-ai

Join Slack

how i can control model language in sts setup

# ask-ai

better-sandwich-65834

12/18/2025, 7:20 AM

how i can control model language in sts setup

tall-airline-16756

12/18/2025, 7:20 AM

Thanks for reaching out, Aleh. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

tall-airline-16756

12/18/2025, 7:20 AM

Do you mean STT (speech-to-text)? If so, here’s how to control the STT model language. • To set language in the model descriptor: specify

provider/model:language

. Example:

stt="deepgram/nova-2:es"

(Spanish). • To set language with the

STT

config object:

Copy code

from livekit.agents import inference

  session = AgentSession(
      stt=inference.STT(
          model="deepgram/nova-2",
          language="fr"
      ),
  )

• For unknown/multi-language audio: ◦ Use

language="multi"

with multilingual models (e.g., Nova-2/Nova-3). ◦ Or use

stt="auto:language-code"

to let LiveKit pick the best model. • To change language at runtime:

Copy code

if self.session.stt is not None:
      self.session.stt.update_options(language="es")

Supported languages depend on the provider/model (e.g., Deepgram Nova-2, ElevenLabs Scribe V2, Cartesia Ink). If you meant something other than STT, tell me which system or give an example of your setup (code snippet or provider/model) and I’ll give exact steps. Sources: Speech-to-text (STT) models | LiveKit Docs | ElevenLabs Change Language | LiveKit docs

better-sandwich-65834

12/18/2025, 7:21 AM

no, i use gpt-realtime in text modality + tts, no stt

tall-airline-16756

12/18/2025, 7:21 AM

Thanks for clarifying — you're using GPT‑realtime in text mode with TTS (no STT). Here’s how you control language and model behavior in that setup. • Control response language via prompts/instructions: ◦ Set persistent system instructions on the Realtime model to force or prefer a language. Example:

instructions="Respond in French."

• Or include explicit per‑request instructions in the conversation messages to switch languages dynamically. • Control model selection: ◦ Use the desired Realtime model when creating the

RealtimeModel

instance (the model choice determines generation behavior). • TTS language/voice: ◦ Choose a TTS provider/model that supports the target language (e.g.,

cartesia/sonic-3

) and set its language/voice options as required. TTS must match the language you instruct the LLM to produce. • Dynamic switching at runtime: ◦ Update model

instructions

or pass language context per request to change language on the fly. ◦ Then call your TTS with the matching language/voice. If you want, tell me: • which Realtime model you’re using, • which TTS provider/model, • and whether you need automatic language detection or explicit control per message — and I’ll give exact code snippets for your setup. Sources: Realtime model plugins | LiveKit Docs | Gemini Live API plugin | LiveKit docs

2 Views

Open in Slack

Previous Next