Use silero-vad PyPI package for offline loading and robust streaming fallback

Disable previous text conditioning and prevent ngram repetitions to eliminate repeat loops

Optimize VAD threshold to 0.6 and add a 5-second forced transcription fallback

Reduce beam_size to 2 to optimize transcription response time on free tier CPU

Fix VAD audio accumulation buffer logic to prevent zero-padding speech chunks

Trust silero-vad repository to bypass interactive prompt

Deploy Yoruba ASR backend (code only)