Information about filtering low frequencies (AUD-1176)

[WilliamTrdjmn]

Hi,

I'd like to know how you filter low frequencies from wavfile. I work on a similar projet and I don't find any information about how making an efficient bandpass filter for low frequency bands (25Hz to 125 Hz for instance) without resampling. Because I'm doing audiostreaming block by block, and I need to keep initial conditions of every filter to keep continuity between each block. So if I resample my signal and filter it in low frequencies and then resample it again to keep a signal sampled at 44100 Hz, I loose initial conditions and so continuities. And that means ''clicks'' between each block (typically 2048 samples).

[jason-mao]

Hi @WilliamTrdjmn， I am not sure I have got your exactly means. Are you want to filter the 25-125Hz audio stream ?

[WilliamTrdjmn]

Hi,

Yes it is. Actually, using scipy and pyaudio, I filter in third octave band an input audio signal block by block with lfilter_zi to keep the delay between each block. For medium and high frequencies, there's no issue of continuity, no «clic». But when I filter low frequencies I need to downsample the signal before filtering to keep a good filter response and stability, then I filter with Fs=Fs/downsampling_factor, and then I need to upsample this signal to sum it with over filtered signals to build the entire filtered signal. So, this method is well for filtering but I also need to stream it in real time and this doesn't keep continuity between each block. I think it's resampling step but I'm not sure. Do you know a good method to filter low frequencies (25Hz-125Hz) with Fs=44100Hz audiostream using python/pyaudio ?

Thank you a lot !

Best regards, William

Le jeu. 17 oct. 2019 à 10:04 PM, maojianxin notifications@github.com a écrit :

Hi @WilliamTrdjmn https://github.com/WilliamTrdjmn， I am not sure I have got your exactly means. Are you want to filter the 25-125Hz audio stream ?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/espressif/esp-adf/issues/257?email_source=notifications&email_token=ALTYD67R2IK74NWHQXKMH33QPEKTZA5CNFSM4I6KL2D2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEBSGOPY#issuecomment-543450943, or unsubscribe https://github.com/notifications/unsubscribe-auth/ALTYD65N6BWH6U7HK4U64Y3QPEKTZANCNFSM4I6KL2DQ .

[houhaiyan]

Hi William I need confirming somthing. first you need band-stop filter, then you need resample (upresample or downresample)? Or first you need resample (upresample or downresample), then you need band-stop filter? If it is ture, you can add band-stop filter element before or after resample elment. And also, we can make special filter which include band-stop filter for you. Best regards,

[WilliamTrdjmn]

Hi @houhaiyan,

In fact, I'm doing these steps (only for low frequencies 25Hz to 160Hz) :
1) downsampling audiosignal with lowpass filter for anti aliasing (using scipy.signal.decimate()) 2) filtering downsampled audiosignal (using IIR chebychev type 2 and scipy.signal.sosfilt() and scipy.signal.sosfilt_zi()) 3) upsampling filtered audiosignal (using scipy.signal.resample()) Comment : there's some clics and discontinuity when I stream it (block by block using pyaudio and callback method)

and for higher frequencies (200Hz to 20kHz): 1) filtering audiosignal (using IIR chebychev type 2 and scipy.signal.sosfilt() and scipy.signal.sosfilt_zi()) Comment : there's no clics and discontinuity when I stream it (block by block using pyaudio and callback method)

Is it better for your understanding ?

[houhaiyan]

@William Hi William Yes I see the step. Is the 25Hz to 160Hz data important for you? CB-II is low-pass filter. And it can make you 25Hz to 160Hz data reserved. If you want reserved the 25Hz to 160Hz data, i think it is ok. If you is to get rid of the 25Hz to 160Hz` data, I think your scheme is error. you should use high-pass filter or band-stop filter. Best regards,

[WilliamTrdjmn]

It's a third octave band bank filter so there're only bandpass filters

[houhaiyan]

octave filter is to select band data which you wanted . And to deal with those data. And the center frequency designed as you wanted. You can use down-resample to improve the precise. @WilliamTrdjmn

[WilliamTrdjmn]

And so ? I need to filter low frequencies during the audiostream and the down-resample step add a delay or something which induces «clic» between each block and in the output audiosignal ..

[houhaiyan]

@WilliamTrdjmn I am no sure. I don't known your need (give up the 25Hz to 160Hz data? Or save the 25Hz to 160Hz data to deal with? You should answer my question.) So I don't known whether you should filter the low frequencies. I think you should follow your need. But in audio stream low frequencies data is important. And octave filter is similar with band-pass filter.

I need to filter low frequencies during the audiostream and the down-resample step add a delay or something which induces «clic» between each block and in the output audiosignal what is clic? If your clic is CLC, in our project, the resample has included the CLC.

[houhaiyan]

sorry, I made a writing error.
If your clic is CIC, in our project, the resample has included the CIC.

[WilliamTrdjmn]

What do you mean CIC ? And yes I need to deal with low frequencies data (adjust magnitude in fact)

Le lun. 28 oct. 2019 à 10:14 PM, houhaiyan notifications@github.com a écrit :

sorry, I made a writing error. If your clic is CIC, in our project, the resample has included the CIC.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/espressif/esp-adf/issues/257?email_source=notifications&email_token=ALTYD6ZJNEZVIU77ITPQHC3QQ6MBBA5CNFSM4I6KL2D2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOECO73JI#issuecomment-547224997, or unsubscribe https://github.com/notifications/unsubscribe-auth/ALTYD65UMAIOSGA2K2WIABDQQ6MBBANCNFSM4I6KL2DQ .

[houhaiyan]

«clic» what's meaning?

[vikramdattu]

@WilliamTrdjmn you might want to take a look at equalizer example, if haven't already. https://github.com/espressif/esp-adf/tree/master/examples/audio_processing/pipeline_equalizer

It is time domain IIR band pass filter. Setting appropriate gain values in set_gain[] array in equalizer_example.c should get you needed outcome.

[jason-mao]

This topic has become inactive so I'm going to close the issue. Please reopen this if you have any questions or need any further assistance.