https://livekit.io logo
Join Slack
Powered by
hi, how can i set the input language for gemini re...
# ask-ai
p
hi, how can i set the input language for gemini realtime conversations (this is for the input audio transcription)
t
Thanks for reaching out, adit-kulkarni. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
Thanks — you're asking how to set the input language for audio transcription in Gemini Realtime Conversations. You set the input language on the STT (speech-to-text) provider you use — not on the Gemini model itself. Pick a supported STT plugin (AssemblyAI, Deepgram, OpenAI, etc.) and set its 
language
parameter to the desired language. For example: • AssemblyAI example:
Copy code
session = AgentSession(
  llm=google.realtime.RealtimeModel(model="gemini-2.5-flash-native-audio-preview-09-2025"),
  stt=assemblyai.STT(
    model="universal-streaming-multilingual",
    language="es"  # Spanish input
  ),
  vad=silero.VAD.load()
)
• Deepgram: use a multilingual-capable model (e.g., 
nova-3
) and set 
language
to a code like 
fr
, 
es
, or 
multi
for automatic language detection. • OpenAI: use the transcribe-capable model and set its 
language
parameter to the desired code. If you tell me which STT provider you're using and which language (or if you want auto-detection), I’ll give the exact config snippet for that provider. Sources: AssemblyAI plugin guide | LiveKit docs | Deepgram STT | LiveKit docs | OpenAI STT plugin guide | LiveKit docs | Gemini Live API plugin | LiveKit docs
p
ah so for interacting with the gemini realtime model i can use any STT model? i dont have to use gemini?
t
Short answer: yes — you can use any compatible STT provider; you don't have to use a Gemini-specific STT. Gemini Realtime accepts text (or transcribed text) as input, so you can run audio through whatever STT model/plugin you prefer (AssemblyAI, Deepgram, OpenAI, etc.), set the language there, then pass the resulting transcript to Gemini. If you want auto language detection, pick an STT model that supports it (e.g., Deepgram 
multi
or AssemblyAI multilingual). Which STT provider are you planning to use? I can give the exact config snippet for that provider.
p
just to be clear, this is for the realtime, audio-to-audio gemini realtime model? im not sure what you mean by passing transcripts to gemini?
t
You're right to double-check — here's the short, clear answer and next steps. Acknowledgement: You're asking if this applies to the realtime audio-to-audio Gemini model and what we mean by "passing transcripts." Answer: • Gemini Live (the realtime audio-to-audio model) accepts audio input and returns audio output; it performs transcription internally. You do NOT have to use an external Gemini-specific STT. • You do not need to manually pass transcripts to Gemini in normal audio-to-audio mode — Gemini transcribes the audio itself and uses that for understanding and response. • Use an external STT only if you want realtime transcripts independently (for UI, logging, filtering) or to override turn-detection/voice-activity behavior. If you want to force language behavior: • You cannot set a direct “input_language” parameter for Gemini Live; it auto-detects language. • You can influence outputs by providing system instructions (e.g., “Respond only in English”) or use an external STT that you configure for a specific language and feed those transcripts into your app logic. Quick question to help further: do you want Gemini to handle everything (simpler audio-to-audio), or do you need external realtime transcripts / forced-language transcription? Sources: Gemini Live API plugin | LiveKit docs
p
well all i want is to use the gemini audio-to-audio model but make sure the transcripts are in a specific language that i set
t
Thanks — I understand you want Gemini audio-to-audio but guarantee transcripts in a specific language. Short answer: • Gemini Live (audio-to-audio) auto-detects and transcribes audio internally and does NOT expose a direct parameter to force the input transcription language. • If you need guaranteed transcripts in a specific language, use an external STT provider configured to that language and capture the transcript yourself; Gemini can still handle audio-to-audio for responses while you use the external transcripts for logging, UI, or verification. Two practical options: • Keep it simple (Gemini-only): let Gemini handle audio-to-audio. You cannot force its internal transcription language, but you can influence responses with system instructions (e.g., “Respond only in English”). • Guarantee transcript language: attach an external STT (AssemblyAI, Deepgram, OpenAI transcribe, etc.) and set its 
language
parameter to the language code you want. Use that provider for transcripts while still sending raw audio to Gemini Live for audio responses. If you want, tell me: • Which STT provider you prefer and the target language code (e.g., 
en
, 
es
, 
fr
), and I’ll give the exact config snippet for integrating that STT alongside Gemini Live.
6 Views
Open in Slack
PreviousNext
Powered by