How to find Frequencies to filter?

Hello all,

I am trying to find and determine what frequencies need filtering. I have an audio signal with a hiss noise at the first 2 seconds of my signal, the hiss has like 4 levels of hiss and each level increases and suddenly stops. I am trying to filter out that hiss using either FIR or IIR (haven't done a deeper research to decide which one to use, Probably FIR [but open for guidance]). My question is where do I look for my frequencies? I have the time signal, fft, spectogram, log scale.

Logarithmic Scale

THANK YOU IN ADVANCE

Answer 1

[Mathieu NOE]

hi

seems you need a low pass filter with a cut off frequency around 2.5 kHz - order to be defined / tested

you should apply it only on the first 1.7 s where the hiss is present

of course the LP filter will also reduce useful infos of the audio signal in that portion of the signal.

Just in case it might further help you , I share my code for FFT / specgram analysis.

I'd suggest to do an averaged pwelch spectrum of the first 1.5 s to see the slope of the hiss noise , to determine which LP order would be needed to flaten or get even a negative slope of the filtered signal.

[Mathieu NOE]

hello

could you please share your wav file ?

tx

[Luis Beltran]

I have attached a Zipfile with my matlab codes I've done so far. .wav file is in the zip folder also. Let me know if you were able to opened it.

Thank you

[Mathieu NOE]

I just tried to reduce the hiss by doing a 5th order bandpass filter (fc = 500 - 2500 hz) on the first 1.68 s of the record

this is how it sounds

[Luis Beltran]

The clip sounds good Mathieu

To achieve this clean clip you used GoldWave? I know that music editors are best for music editing but is it wrong what I am trying to do? Could I achieve a similiar clip as the one you provided using MATLAB? I can create a bandpass filter if I refer back to my notes but could you send a link that provides information on how you got the 5th order and the cut off freq? Or could you explain? I've watched some videos on youtube and looked through google but still confuse.

Thank you for your help.

[Mathieu NOE]

hello Luis

I replicate the bandpass filter on the initial segment of the audio file in the matlab code

I added a second spectrogram plot after the filtering to see the effect

listening to the file generated by matlab, very similar to Goldwave

see attachement

[Luis Beltran]

Good Evening Mathieu,

Sorry for my late response, I was doing some progress on a analog project.

I've review the attachement you provided and the clip sounds goods from MATLAB. I am still confuse on how to determine the frequencies, was it just by inspection?

Also some of the functions used on the attachement I saw them today in class. So the methods provided are using Fourier Methods and Filtering right? We are barely experimenting with this stuff in class and doing the project and relating it to the class content seems to be making sense

[Mathieu NOE]

hello Luis

yes indeed , I selected the filter just by looking at the spectrogram frequency amplitudes (colors) during the first 1. 7s

as it was quieter in the 500 / 2500 hz, no need to reduce sound in that frequency range

you see the amplitudes ges up above 2500 Hz, and for the human ear , the max sensivity is between 1 and 5 kHz, so if you have a noise spectrum with larger amplitude above 2.5 kHz , you will hear it, and that's what has to be tamed down.

you could alos completely remove that section of the audio recording, it's even more efficient !

yes , all the methods we are using here are based on fft for spectral analysis (averaged or time dependent) , and we use all available digital filters to shape the spectrum of a signal

the demo files in matlab are very usefull to learn and pratice those fields. You never stop to learn new stuff

attached other examples of usefull digital filters

[Luis Beltran]

Thank you Mathieu you've been of extremely help!

I was working on some analog projects, but I am finishing my filtering and I found some webinars on Spectral Analysis.

https://www.youtube.com/watch?v=qrU2jsSqWD8&t=63s

[Mathieu NOE]

you're welcome

good luck for the future !