Instruction.txt

1. Speaker embedding

1.1 Generate pattern
python -m Speaker_Embedding.Pattern_Generate [options]

option list:
-vctk <path>		Set the path of VCTK. VCTK's patterns are generated.
-ls <path>		Set the path of LibriSpeech. LibriSpeech's patterns are generated.
-vox1 <path>		Set the path of VoxCeleb1. VoxCeleb1's patterns are generated.
-vox2 <path>		Set the path of VoxCeleb2. VoxCeleb2's patterns are generated.

1.2 Set inference files path while training for verification
Edit 'Speaker_Embedding_Inference_in_Train.txt'

1.3 Run
python -m Speaker_Embedding.Speaker_Embedding


2. Mel to spectrogram

2.1 Generate pattern
python -m Taco1_Mel_to_Spect.Pattern_Generate [options]

option list:
-vctk <path>		Set the path of VCTK. VCTK's patterns are generated.
-ls <path>		Set the path of LibriSpeech. LibriSpeech's patterns are generated.

2.2 Set inference files path while training for verification
Edit 'Mel_to_Spect_Inference_in_Train.txt'

2.3 Run
python -m Taco1_Mel_to_Spect.Taco1_Mel_to_Spect


3. Multi speaker TTS

3.1 Generate pattern
python Pattern_Generate.py [options]

option list:
-lj <path>		Set the path of LJSpeech. LJSpeech's patterns are generated.
-vctk <path>		Set the path of VCTK. VCTK's patterns are generated.
-ls <path>		Set the path of LibriSpeech. LibriSpeech's patterns are generated.
-tl <path>		Set the path of Tedlium. Tedlium's patterns are generated.
-timit <path>		Set the path of TIMIT. TIMIT's patterns are generated.
-all		All save option. Generator ignore the 'Use_Wav_Length_Range' hyper parameter. If this option is not set, only patterns matching 'Use_Wav_Length_Range' will be generated.

3.2 Set inference files path while training for verification
Edit 'Inference_Sentence_in_Train.txt'

3.3 Run
python MSTTS_SV.py