UNIFIED MEMORY AUDIO IN TEXT TOKENS CONTEXT AUDIO OUT STT (WHISPER) LLM (GEMMA) TTS (CHATTERBOX) 0.5s warm-up Markdown Stripping RMS amplification