Topic: mel-log-spectrum-approximation Goto Github

👇 Here are 1 public repositories matching this topic...

bayuwira / kendang-tunggal-classification-using-backpropagation-and-onset-detection
mel-log-spectrum-approximation,Bali has a diversity of arts that has been recognized by the world, where one of the most famous Balinese arts is the Karawitan art, especially the Kendang Tunggal instrument. Notation documentation or more commonly known as music transcription, can make learning a song easier, and in the case of this research, it makes it easier to learn to play the Kendang Tunggal instrument. The first approach method used to document a kendang tunggal song is onset detection. Onset is when the signal experiences an attack period, which helps segment the sound color of the drum instrument. The segmented kendang tunggal sound color classification uses the Backpropagation algorithm with several features of the frequency domain and time domain as a characteristic of the sound color. Then the kendang tunggal song is revived into a synthetic sound with the Mel Spectral Approximation filter. Based on the research, the optimal parameter for drum sound color segmentation with onset detection is the hop size 110 with normalization of the features on its onset detection function. The optimal backpropagation architecture obtained with a learning rate of 0.9, neurons 10, and epoch 2000 produces an accuracy of 60.85%. The synthesis method using the Mel Log Spectrum Approximation can make synthetic sounds similar to kendang songs with an accuracy of 83.33%
User: bayuwira
classification onset sound synthesis drums machine-learning machine-learning-algorithms fastfouriertransform mel-spectrogram zero-crossing-rate-variation