Confirmation behavior: strict (explicit yes only), flexible (implicit OK), auto-proceed (timeout-based)
Seconds to wait before re-prompting for confirmation (3-15)
Default language when detection confidence is too low (e.g., 'fr-FR')
Minimum confidence percentage (50-100) to accept detected language
Honor user requests to stay in a specific language (e.g., 'rester en français')
Language/locale code (e.g., 'fr-FR', 'nl-NL', 'de-DE')
Enable automatic language detection
Mapping of language codes to default voice IDs (e.g., {'fr-FR': 'Lea', 'nl-NL': 'Lotte'})
Audio padding before speech in ms (only used when vad_type=server_vad)
Silence duration in ms before end-of-speech (only used when vad_type=server_vad)
Semantic VAD eagerness level (only used when vad_type=semantic_vad)
VAD threshold (0.0-1.0, only used when vad_type=server_vad)
OpenAI STT VAD type: semantic_vad (AI-based) or server_vad (silence-based)
Consecutive language disagreements needed to trigger re-detection (3-10, default 5)
Enable Whisper language detection (only used when stt_language=auto)
Whisper language (auto|en|fr|nl|de)
Provider-specific STT model (OpenAI: whisper-1; Groq: whisper-large-v3, distil-whisper-large-v3-en)
Speech-to-text provider: aws|openai|groq
List of supported language codes for detection (e.g., ['en-US', 'fr-FR', 'nl-NL'])
Minimum milliseconds between voice switches (5000-60000, default 15000)