RoxasYTB's picture
Upload README.md with huggingface_hub
db4d659 verified
|
Raw
History Blame Contribute Delete
978 Bytes
metadata
configs:
  - config_name: default
    data_files:
      - split: train
        path: data.zip
tags:
  - audio
  - voice
  - portal
  - piper-tts
  - ljspeech
license: cc-by-4.0

Wheatley — English (en)

LJSpeech dataset of Wheatley (en).

  • 827 pairs
  • 44.1kHz mono 16-bit PCM WAV (original wiki quality)

Piper TTS Training (High Quality on T4 GPU)

Preprocessing (downsample to 22.05kHz)

python3 -m piper_train.preprocess \
  --language en-us \
  --input-dir ./wheatley_en \
  --output-dir ./train_wheatley_en \
  --dataset-format ljspeech \
  --single-speaker \
  --sample-rate 22050

Training (Kaggle T4 16GB)

python3 -m piper_train \
  --dataset-dir ./train_wheatley_en \
  --accelerator 'gpu' \
  --devices 1 \
  --quality high \
  --batch-size 16 \
  --validation-split 0.0 \
  --num-test-examples 0 \
  --max_epochs 10000 \
  --resume_from_checkpoint /path/to/checkpoint.ckpt \
  --checkpoint-epochs 1 \
  --precision 32 \
  --max-phoneme-ids 400