walston/joycent-large_v3_turbo_grl

Joycent trained with WhisAID accent embeddings

This Joycent Mandarin accent TTS acoustic model was trained using accent embeddings extracted by walston/whisaid-large_v3_turbo-grl. The released checkpoint is epoch 100.

• Accent encoder: walston/whisaid-large_v3_turbo-grl
• Vocoder: walston/joycent-vocoder

Download

from huggingface_hub import hf_hub_download

checkpoint_path = hf_hub_download(
    repo_id="walston/joycent-large_v3_turbo_grl",
    filename="grad_100.pt",
)

Pass the path to joycent/inference_joycent.py with --acoustic-checkpoint. Full synthesis also requires the Joycent vocoder.

Checkpoint

• Epoch: 100
• Acoustic model: Joycent / Grad-TTS
• Accent embedding dimension: 256