Moved Speech-to-Text processing onto Gemini 3.5, improving transcription accuracy and reducing hallucinated transcripts on silent or non-speech audio.