for realtime audio stream difference in stft and df.analysis

[DonkeyHang]

I want to process by audio stream, for example, when i process Realtime in stream mode, i cant aldeady had enough samples, i only can capture 10ms(480 samples) and need to get 480 samples after effects, but the scripts was input all of data, even if it was split for any frames, i input 960 samples to stft process used hann window, and overlap 480 samples, and the first time i pad 480 sample zero,

but the question come, the result is difference with df.analysis

so could you told me how the df.analysis process mode?

thx

[DonkeyHang]

ok，the window function is vorbis, i do a lot of test and thus spec result was look fine but when uesd stream mode, it seem to pad block zeros, i'll do more test to make sure result accuracy

[DonkeyHang]

use torch.fft.rfft and multi vorbis window can get same result, but in window part, we need do some diff in index