使用RNN模型作語音訊號的除噪

使用RNN模型作語音訊號的除噪

 O​riginal Voice Clip:錄下的人聲語音訊號

 O​ffice Noise:錄下的辦公室背景噪音

 Mixing:前２項混合而成的訊號

 After RNN:透過RNN模型除噪後的語音訊號波形

 By Directional Mic in F15K:筆電上指向型麥克風的所錄下除噪後的波形

結論：透過RNN模型除噪的功能，具有近似指向型麥克風的除噪功能。也就是以軟體處理的技術來取代實體裝置。

 From lower right 2 pictures, voice clip mixed with background noise was restored well in comparison to audio waveform recording by directional mic and original voice clip waveform.

Ref:
https://people.xiph.org/~jm/demo/rnnoise/
https://hacks.mozilla.org/2017/09/rnnoise-deep-learning-noise-suppression/
https://github.com/xiph/rnnoise

mic denoise rnn
        audacity:
        import -> raw data -> Signed 16bit/Little endian/one channel, 48000
        rnn
        frank@frank-GL753VD:~/1T/back0529/mic
        import format: Signed 16 bit mono format
        frank@frank-GL753VD:~/1T/back0529/mic
        ~/1T/back0529/mic/RNNtest/F15K/tsai1.raw  ---->recorded by 2 mic (directional mic)
        ~/1T/back0529/mic/RNNtest/AS3EA/OUTtsai1.pcm  ---->after RNN
        ~/1T/back0529/mic/RNNtest/AS3EA/tsai1.pcm  ----> after Mixing