ArtC

Hello, I would like to present the results of my last test. I cleared a speech sound from the noise. In the image below, you can see what the spectra of these speech sounds look like. Some might say that noise removal is nothing fancy, but in my case, I used an algorithm that I intend to use to reconstruct audio speech from conferencing applications such as MS Teams, Zoom, or Webex to recover what was lost during the compression and transmission process.. And I'm pleased with myself because I did it in less than the duration of the speech, so it's very likely that I'll be able to do this operation in a real-time application. Follow me, I will post some examples for you to listen to in my next post.

https://artc-project.blogspot.com/2023/06/real-time-application-is-possible.html

Recently, I conducted preliminary tests of algorithms that have a good chance of solving the problem faced by this project. In my tests, I managed to get processing time in the same order of magnitude as the recording time. I find it very promising because I see the possibility of improving performance to the extent that it works in real-time. At the moment I am preparing to test the algorithm when it comes to sound quality, or how to improve its quality

It turns out that problems with comprehension and listening comfort also occur on VOD platforms. This seems to be a fairly common problem. This sounds like another potential application
https://www.youtube.com/channel/UCLXo7UDZvByw2ixzpQCufnA/community?lb=Ugkxk4WyjJPTbkoDXCB1TCEvztPvQ-CCAIEi

After a long break, I came back to this project. I figured the first thing I needed to do was objectively compare the original audio with the audio recorded by the microphone and the audio from MS Teams.
So I equalized the volume of these audios, and I equalized them together in the time domain. I found a correlation between the original audio and the microphone audio to align them together.
I'm particularly interested in the frequency domain, so I've made an effort to display the spectrograms of these signals. My goal was to visualize the details of the signals and do the FFT manually to fully understand how the signals were processed.

https://artc-project.blogspot.com/2023/04/after-long-break-i-came-back-to-this.html

Today assembled my test stand. There are two low-quality microphones and quite nice speakers. I'm sure that some of you would claim that the speakers are garbage as well but in my opinion, it's good enough. I can listen to speeches with some comfort, but I can't say that about the output of that mics. So, it seems to be working as I desired :D

https://artc-project.blogspot.com/2022/12/today-assembled-my-test-stand.html

I compared referential speech with MS Teams sound. The difference is great. You can hear it, you can see it. It turns out that MS adheres to the principle of "eat the meat, leave the potatoes", They don't care about user experience, the transfer usage is the only thing that matters.
https://artc-project.blogspot.com/2022/12/i-compared-referential-speech-with-ms.html

I managed to visually compare to signals. The first signal is referential speech recording and the second is speech disturbed by microphone imperfections. Now is the time for deeper analyses and more demanding tests that will show where the issue is, and how to solve it.

During my daily, I came up with idea that better quality of voice could help me focus on the content of the meeting. Then I decided to investigate that issue and create a tool to recover some voice properties and make that voice pleasant to listen to