Audio de-noising using Python (Wavelets)

-

What is audio de-noising?

Audio de-nosing is a process of removing noise(unwanted sound/signals from audio). You might have heard of noise cancellation headphones, these remove all the noise from the audio.

 This process has various applications in media creation and learning. Also, de-noising is not only done on audio but also done on images. There are various methods to do this. We will see 2 methods and implement one.


Method 1: FFT (Fourier Transform)

   There are multiple different implementations of FT: FFT (fast FT), STFT (Short time FT). Also, there are inverse for each of them IFT respectively.

So what is it? In short, FFT converts signals from one form to another, in audio, the time-series signal is converted to the frequency domain.

More shortly, the dominant audio (your voice) will have more power than the non-dominant(noise).

The following video will explain (a really good video): 

Denoising with FFT (in Python) 


Method 2: Wavelets (faster than FT)

   These are complicated to understand but in short, it will decompose the signal into levels and windows, resulting in a window where the noise is not there, and only the important stuff exists.

The following video from the same person as above:

Wavelets and multi-resolution analysis 


Implementation

Most of you like me will jump directly into coding. I will only explain a few things here.




Requirements: 

1. pywt

2. soundfile 

3. numpy

4. tqdm (just a loading bar, you can remove it)

 

The function 'mad : median absolute deviation', it is available in statsmodel package but only for one function I didn't want to install a package.

 

 

  • Using soundfile.info we get all information about the file like frequency rate(14400Hz), duration, and other metadata.
  • Also, you can calculate the duration of the audio by (len(data) / rate) gives the time in seconds.
  • We create a numpy array to store the cleaned signal
  • Loop through the blocks of the file, some files can be big and take up a lot of memory, so in this approach, I read the file in blocks, you can change the block size (2nd parameter) to suite you.
  • If the shape is greater than 1 ( multi-channel audio), select only the single-channel (mono)
  • wavedec : wavelet decomposition, this calculates the windows with specified level and wavelet method. 
  • then calculate the median average deviation
  • get a threshold value for noise
  • apply the threshold value to all the coefficients
  • finally waverec : recomposition ( make sure you use the same wavelet and mode for both decompositions)
  • save our file and see the result.


There are many things to do with audio, but I wanted to share this application, cause I couldn't find a great one. 

Source code : Audio-Denoising


Comment below for any queries, mistakes, anything ;)

Popular posts from this blog

Using FFMPEG to trim your videos

-
 There are multiple methods to trim or clip videos with the help of FFMPEG, but sometimes looking at the documentation can confuse lots of us, and we cannot derive the output we want. And the examples on the documentation are simpler than what we want to do. So I will show few methods to combine, trim or clip video files using FFMPEG. I won't be going through installation process and will assume that you already have it installed. 1. Using timestamps to clip the video 🕐   ffmpeg -i video.mp4 -ss 00:00:00 -to 00:10:00 -c copy output.mp4   This command will copy the clip from 0 sec to 10 min without re-encoding, since we specify not to re-encode. Break down of command:      -i  : input video file     -ss : starting timestamp you can define in hh:mm:ss or just secs     -to : ending timestamp, this is the timestamp of the video     -c copy : codec is copy, which doesn't require re-encoding   If you want to copy the video from some other time to some duration    ffmpeg -ss 00
Read more

Converting HTML5 Game for Android | 2 Methods

-
HTML5 games are easy to develop and use. It has great graphics and processing capabilities. Many may wonder to convert their web apps compatible with Android. So here are two methods, which will walk you through Android development for making your web app compatible with your favorite platform. Method 1: Hosting the Game  There are various sites where you can host your HTML game for free and also earn from it. After hosting it take the link of the Game use it to make your Android game. 1. Create a new Android Project (obviously) 2. Add the following to the '.XML' file of the layout. In the above snippet, a WebView is created which takes up the whole screen. And a ProgressBar which is by default round and will appear when the page is loading. Till now there won't be any errors 😅 3. In your MainActivity.java, start by typecasting the WebView and ProgressBar. 4. Add web settings and enable JavaScript. The above code creates some basic Web Sett
Read more