frame length about porcupine

i have removed wave from the imports and just using soundfile now
everything is working well

[kenarsa]

correct. It accepts 512 samples per frame. You can buffer the audio and pass it to engine every other frame.

maybe im not understanding samples or frames correctly

so the stream comes in as a wave file.. so when i look at the wav info i get
_wave_params(nchannels=1, sampwidth=2, framerate=16000, nframes=256, comptype='NONE', compname='not compressed')

so i set nFrames = 256

and in code
b = io.BytesIO(msg.payload)
audio = wave.open(b, 'rb').readframes(nFrames)
result = porcupine.process(audio)

and result never fires the hotword being found

thought maybe the audio code is bad.. so added pyAudio to listen to the stream
pa = pyaudio.PyAudio()
stream = pa.open(format=pyaudio.paInt16, channels=1, rate=16000, output=True, frames_per_buffer=nFrames)

and then the code is
b = io.BytesIO(msg.payload)
audio = wave.open(b, 'rb').readframes(nFrames)
stream.write(audio)
result = porcupine.process(audio)

the audio playing is crystal clear so its not bad code converting the audio.. and so added the code print(len(audio)) before the stream write.. it outputs to console 512. 512. 512 .... << this is the 512 you refer to porcupine liking??

as you said porcupine like 512.. so i changed the stream to 512.. and ran again
_wave_params(nchannels=1, sampwidth=2, framerate=16000, nframes=512, comptype='NONE', compname='not compressed')
and changed nFrames = 512

again the audio playing from pyAudio is crystal clear.. now the console logs 1024. 1024. 1024..
but again porcupine never picks up the hotword

any idea what might be wrong? im not sure if the sample width being 2 and nFrames 512 makes the console len print 1024? but with the stream at 256 or 512.. and the len outputting 512 or 1024 porcupine never fires

so i changed using wave to soundfile
replaced the code for opening the wave with
audio, sample_rate = soundfile.read(b, dtype='int16')

the len of audio console prints 512.. good :)
porcupine fires on the hotword.. good :)

pyAudio does not sound good at all.. its all choppy
nFrames = 512
pa = pyaudio.PyAudio()
stream = pa.open(format=pyaudio.paInt16, channels=1, rate=16000, output=True, frames_per_buffer=nFrames)
...
stream.write(audio)

sounds choppy.. not clear like it was using the wave lib to open it.. but porcupine works with soundfile whereas it does not with wave.. using the same code

update
changing the audio to bytes for pyaudio now sounds clear again and not choppy
stream.write(bytes(audio))

[kenarsa]

Could it be an endianness issue?