Denoised audio using LSTM

features: log fft 512, lstm 1024 units

dirty = dirty signal

predi = predicted signal using LSTM

spectrum: from top to bottom:
1 predicted spectrogram,
2 resynthesis spectrogram based on dirty signal complex spectrum
3 clean signal
4 dirty signal

File #0

dirty

predi

File #2

dirty

predi

File #3

dirty

predi

File #1

dirty

predi

File #4

dirty

predi

File #6

dirty

predi

File #5

dirty

predi