Wavenet vocoder training and evaluation

Preprocess

python preprocess.py liepa_multi ./db/Garsynas/ ./db/proc/Garsynas --preset=presets/liepa_multi.json

Resulting train.txt containing list of utterance transcriptions for corresponding speaker.

Train

python train.py --data-root=./db/proc/Garsynas --preset=presets/liepa_multi.json

Evaluate

python evaluate.py checkpoints/checkpoint_step000590000_ema.pth \
  ./generated/Garsynas_multi_eval --data-root=./db/proc/Garsynas/ \
  --preset=presets/liepa_multi.json --output-html --num-utterances=1

Fine tune

python train.py --data-root=./db/proc/Garsynas/ \
  --checkpoint-dir=checkpoints/fine_tune \
  --restore-parts=checkpoints/checkpoint_step000590000_ema.pth