mirror of
https://github.com/storytold/storyteller-ml.git
synced 2026-10-28 09:08:54 +00:00
ZDisket
456fc04e63
- Tacotron2 based on https://github.com/keonlee9420/Comprehensive-Tacotron2 - Removed DeepSpeaker and Tensorflow dependency - Added ARPA function - Replaced guided attention loss with Convolutional Attention Consistency
33 lines
825 B
YAML
33 lines
825 B
YAML
dataset: "VCTK"
|
|
|
|
path:
|
|
corpus_path: "/mnt/nfs2/speech-datasets/en/VCTK-Corpus-0.92"
|
|
wav_tag: "mic1"
|
|
wav_dir: "wav48_silence_trimmed"
|
|
txt_dir: "txt"
|
|
preprocessed_path: "./preprocessed_data/VCTK"
|
|
|
|
preprocessing:
|
|
sanity_check: False # if Ture, then preprocess a single data and quit
|
|
speaker_embedder: "DeepSpeaker" # support 'none', 'DeepSpeaker'
|
|
val_size: 512
|
|
text:
|
|
text_cleaners: ["english_cleaners"]
|
|
language: "en"
|
|
audio:
|
|
sampling_rate: 22050
|
|
max_wav_value: 32768.0
|
|
skip_len: 14848
|
|
trim_top_db: 23
|
|
silence_audio_size: 5
|
|
pre_emphasis: False
|
|
stft:
|
|
filter_length: 1024
|
|
hop_length: 256
|
|
win_length: 1024
|
|
mel:
|
|
n_mel_channels: 80
|
|
mel_fmin: 0
|
|
mel_fmax: 8000 # please set to 8000 for HiFi-GAN vocoder, set to null for MelGAN vocoder
|
|
normalize: False
|