Week 6 — Audio Emotion Recognition System (Part IV)

Hello everyone ! This is our last blog about our machine learning project. Last week, we shared the results of methods that we use. This week, in addition to them, we share some results too.

Let’s start !

We applied CNN algorithm to our resized grayscaled amplitude images and average result is 0,35. Following image respresents the weights of the first layer..

Then, we tried another classifier to improve our accuracy. We took the mean of mfcc and “chroma_stft”, “chroma_cqt”, “chroma_cens”, “rms”, “spectral_contrast”, “spectral_bandwidth”, “tonnetz”, “zcr” features and applied Decision Tree algorithm. Accuracy is 0,35.

Then, we used XGBoost as classifier. Result of this algorithm is 0,40.

In conclusion, we worked with many algorithms to improve our model’s success. According to all results, best model is CNN with mfcc feature as input.

You can watch the video presentation from here →

Thank you for reading !

Nur Altıparmak — https://www.linkedin.com/in/nur-alt%C4%B1parmak-5722a9146/

Ece Omurtay — https://www.linkedin.com/in/ece-omurtay/

Previous posts:

Week 1 — https://medium.com/bbm406f19/week-1-introduction-557d0143e753

Week 2 — https://medium.com/bbm406f19/week-2-data-analysis-687ec86c0a71

Week 3 — https://medium.com/bbm406f19/week-3-audio-emotion-recognition-system-1e37a4991a48

Week 4 — https://medium.com/bbm406f19/week-4-audio-emotion-recognition-system-part-ii-ba8f24704db5

Week 5 — https://medium.com/bbm406f19/week-5-audio-emotion-recognition-system-part-iii-a4f6ee87f458