LiveKit’s MultilingualModel in the voice agent pipeline is compatible with any STT plugin that can report the language of the user’s speech. This includes over 10 STT providers available in the LiveKit agents framework, such as: • Amazon Transcribe • AssemblyAI • Azure AI Speech • Azure OpenAI • Baseten • Cartesia • Clova • Deepgram • fal • Gladia • Google Cloud • Groq • OpenAI • Sarvam • Speechmatics • Spitch For example, Deepgram STT models like "nova-2" and "nova-3" are commonly used with the MultilingualModel. The key requirement is that the STT plugin must support language settings compatible with the MultilingualModel’s supported languages: English, Spanish, French, German, Italian, Portuguese, Dutch, Chinese, Japanese, Korean, Indonesian, Turkish, and Russian. You can configure the STT plugin with a specific language or use automatic language detection (e.g.,

language="multi"

with Deepgram). Sources: LiveKit turn detector plugin | LiveKit Docs | Speech-to-text (STT) integrations | LiveKit Docs | Voice AI quickstart | LiveKit Docs | Voice AI quickstart | LiveKit Docs | LiveKit turn detector plugin | LiveKit Docs