Original Authors : Dabi Ahn

Original Project URL : https://github.com/andabi/deep-voice-conversion

Thanks to the original Dabi Ahn for his voice conversion project on GitHub: Deep-Voice-Conversion. PPG_Tacotron project is an implementation of the Deep-Voice-Conversion project based on PyTorch. This implementation based on PyTorch improved the training speed of the model to 9x, and the generated speech quality was consistent with the original project

This is a many-to-one voice conversion system, which adopts the speech conversion model proposed by ICME2016 : Phonetic posteriorgrams for many-to-one voice conversion without parallel data training, CBHG module and PreNet module in Tacotron model were used to optimize the model

The main significance of this work is that we could generate a target speaker's utterances without parallel data like <source's wav, target's wav>, <wav, text> or <wav, phone>, but only waveforms of the target speaker. (To make these parallel datasets needs a lot of effort.) All we need in this project is a number of waveforms of the target speaker's utterances and only a small set of <wav, phone> pairs from a number of anonymous speakers.

The model architecture consists of two modules:

1. Net1(phoneme classification) classify someone's utterances to one of phoneme classes at every timestep.
   • Phonemes are speaker-independent while waveforms are speaker-dependent.
2. Net2(speech synthesis) synthesize speeches of the target speaker from the phones.

• Process: wav -> spectrogram -> mfccs -> phoneme dist.
• Net1 classifies spectrogram to phonemes that consists of 60 English phonemes at every timestep.
  • For each timestep, the input is log magnitude spectrogram and the target is phoneme dist.
• Objective function is cross entropy loss.
• TIMIT dataset used.
  • contains 630 speakers' utterances and corresponding phones that speaks similar sentences.
• Over 70% test accuracy

Net2 contains Net1 as a sub-network.

• Process: net1(wav -> spectrogram -> mfccs -> phoneme dist.) -> spectrogram -> wav
• Net2 synthesizes the target speaker's speeches.
  • The input/target is a set of target speaker's utterances.
• Since Net1 is already trained in previous step, the remaining part only should be trained in this step.
• Loss is reconstruction error between input and target. (L2 distance)
• Datasets
  • Target(anonymous female): Arctic dataset (public)
• Griffin-Lim reconstruction when reverting wav from spectrogram.

• python 3.7
• pytorch == 1.5
• librosa == 0.7.2

• sample rate: 16,000Hz
• window length: 25ms
• hop length: 5ms

• Train phase: Net1 and Net2 should be trained sequentially.
  • Train1(training Net1)
    • Run train_net1.py to train and test_net1.py to test.
  • Train2(training Net2)
    • Run train_net2.py to train and test_net2.py to test.
      • Train2 should be trained after Train1 is done!
• Convert phase: feed forward to Net2
  • Run convert.py to get result samples.
  • Check Tensorboard's audio tab to listen the samples.
  • Take a look at phoneme dist. visualization on Tensorboard's image tab.
    • x-axis represents phoneme classes and y-axis represents timesteps
    • the first class of x-axis means silence.

• Window length and hop length have to be small enough to be able to fit in only a phoneme.
• Obviously, sample rate, window length and hop length should be same in both Net1 and Net2.
• Before ISTFT(spectrogram to waveforms), emphasizing on the predicted spectrogram by applying power of 1.0~2.0 is helpful for removing noisy sound.
• It seems that to apply temperature to softmax in Net1 is not so meaningful.
• IMHO, the accuracy of Net1(phoneme classification) does not need to be so perfect.
  • Net2 can reach to near optimal when Net1 accuracy is correct to some extent.

• "Phonetic posteriorgrams for many-to-one voice conversion without parallel data training", 2016 IEEE International Conference on Multimedia and Expo (ICME)
• "TACOTRON: TOWARDS END-TO-END SPEECH SYNTHESIS", Submitted to Interspeech 2017