Code not able to catch user speech well

When speaking in English, it captures in some other language and reply in that. Even I force tts model to use english. Still not working.

same thing here too