mizoru / fastspeech2

TTS Project to train DeepSpeech2

Installation guide

You first run this in your shell:

git clone https://github.com/mizoru/FastSpeech2.git
cd FastSpeech2
pip install -r requirements.txt
wget https://dagshub.com/mizoru/FastSpeech2/raw/be43b4f7d3be88e258e0cef8cdd68d587fff54e7/checkpoint1.pth

Inference

To get the audio for sentences in texts.txt run:

python test.py -m checkpoint1.pth -t texts.txt

Fine-Tuning

To continue training:

python train.py -r checkpoint1.pth

Results

You can find the final inference-time predictions in wavresults.zip

Training Logs

You can find the report on the training here and also look at my WandB logs

Synthesis Parameters

B(eta) stands for pitch and G(amma) for energy