I am performing simple audio recognition using tensor-flow to spot key word or hot word detection. The graph takes the microphone input directly and performs "Audio spectrometer --> mfcc's --> and input to the neural network."

Now i would to like to perform some kind of real time back ground noise reduction or cancellation before inputting the mfcc's into the neural network.

I am not sure what kind of algorithms/python packages that are helpful to reduce the back ground noise and If some one can point me to the pieces of code/knowledge base that can help to filter the mfcc's before inputting into neural network.

  • $\begingroup$ Why do you want to do this? A good classifier is robust against noise. In fact, I would go the other way - use data augmentation to inject noise and train also on noisy samples. This will typically help the classifier become more robust against noise $\endgroup$ – jonnor Jan 23 '19 at 14:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.