Publication date: 28 February 2026
The setup was minimal. A small FastAPI server handles an incoming WebSocket connection from Twilio, which streams base64-encoded μ-law audio packets at 8kHz in ~20ms frames. Each packet was decoded and fed into a Voice Activity Detection model - in my case, Silero VAD.
，详情可参考体育直播
Follow topics & set alerts with myFT
美股三大指数收盘涨跌不一，英伟达跌超5%，市值蒸发1.77万亿元