audio file as input parameter for model.transcribe works well but ndarray-typed parameter captured with sounddevice does not work

[cyflhn]

model.transcribe works well when I use an audio file as an input parameter. But when I use sounddevice to record a period of speech and save the speech result as ndarray and send it directly for model.transcribe , it cannot recognize speech.
But I save the speech recorded by sounddevice as an audio file and then use this file as input paramter for model.transcribe , the speech can be recognized. What is the problem? Is there any specific format requirement for ndarray parameter?

[MahmoudAshraf97]

make sure the array is mono and sampled at 16khz float32

[cyflhn]

mono

what does mono array mean? Could you please give me an example? I am not quite into video technique. Here is my code for recording speech:
recording = sd.rec(int(duration * fs), samplerate=fs, channels=2, device=device_index)

[MahmoudAshraf97]

Mono means single channel
And sr should be 16000

[cyflhn]

Mono means single channel And sr should be 16000

I modify my code according to your suggestion, but still did not work.