search

ibab / tensorflow-wavenet

A TensorFlow implementation of DeepMind's WaveNet paper
MIT License
5.41k stars 1.3k forks source link

Question - WaveNet as feature extractor from audio #287

Open iariav opened 6 years ago

iariav commented 6 years ago

Hi, first, thanks for this implementation of WaveNet!

I'm interested in performing feature extraction from raw audio files. this features will be used for different tasks such as voice activity detection, speech recognition etc.

I read the paper and the issues section here on GitHub but couldn't find an answer to my question - can a WaveNet model be used to extract meaningful features from raw audio? say I have a trained WaveNet model - can I use it on new raw audio data to just extract features without the audio generation part?

thanks in advance..

KyungsuKim42 commented 6 years ago

I think you can just use convolutional layer's feature map. Generation part is not necessary I think.

nowachri-de commented 5 years ago

In this article it is described how WavNet is used to extract features from audio

https://medium.com/@LeonFedden/comparative-audio-analysis-with-wavenet-mfccs-umap-t-sne-and-pca-cb8237bfce2f