WEEK -4 Sound of the City
Team Members: Buğrahan AKBULUT, M.Çağdaş ÇAYLI
We have started to coding part of our project at this week.Firstly, we extract the features from our dataset which includes nearly 15 GB sound.We extract some general sound features those are ;
- Chroma,
- Contrast,
- Mel Spectogram,
- MFCC (Mel-frequency cepstral coefficients),
- Tonal Centroid features.
After trying some of these feature vectors with some learning algorithms ( like Neural Nets, SVM…) we will decide which direction do we go, which features are more appropriate for our problem and how can we improve our results.I have mentioned before that our data set has 10 sound classes, we are planning to add some new classes to our dataset after getting some acceptable results.
We use python librosa library to extract these feature by using raw sounds.
Some Raw Sound Example Graphs
More to read…
Here are some resources about features we used and librosa library, if you are interested in take a look at these links:
- https://dmitri.mycpanel.princeton.edu/tonnetzes.pdf (About Tonal Centroid)
- http://recognize-speech.com/feature-extraction/mfcc (About MFCC)
- http://www.nyu.edu/classes/bello/MIR_files/tonality.pdf (About Chroma)
- https://librosa.github.io/librosa/tutorial.html (librosa tutorial)
