Datasetpublic

RoxasYTB/wheatley_en-ljspeech

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… See the full description on the dataset page: https://huggingface.co/datasets/RoxasYTB/wheatley_en-ljspeech.

sourceHugging Facecc-by-4.0updated 3mo agoView on Hugging Face

0likes14downloads

Dataset Card

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