-
active muting
-
mikolaj filtering
Inspiration
On a more serious note, there are many situations you find yourself where its less about the background noise but more about a specific persons voice that is causing issues. Whether its the loud guest keeping your baby up, whether its your friend who will not stop yapping on the phone late in the evening, or you have a friend called Mikolaj who you like to pick on ;).
What it does
We process raw audio, identify whether or not our target person is being heard, and extract the target from the mix audio for clean, Mikolaj free audio.
Challenges we ran into
There were some clear challenges with pure computation. This model is a demostration of the potential should you have better computers to run these models on, therefore the live demo is not nearly as clean as the processes audio. Onto this, the model with the cleanest output, takes significant compute for a clearer sound, which makes my laptop hot enough to cook an egg.
Accomplishments that we're proud of
It actually works quite well, which is incredibly impressive.
We also needed a way to actually track if it was getting better, so the scoring system, helps plot this statistically.
What's next for Mute Mikolaj
Mute them all. Mwahahahahah.
Built With
- python
- speechbrain
- stepformer
