facebook

/

tts_transformer-en-200_speaker-cv4

like 1

models, cfg, task = load_model_ensemble_and_task_from_hf_hub(
"facebook/tts_transformer-en-200_speaker-cv4",
arg_overrides={"vocoder": "hifigan", "fp16": False}
)

Using the hifigan throws an error. If this is switched to the griffin_lim vocoder the error is resolved.

models, cfg, task = load_model_ensemble_and_task_from_hf_hub(
"facebook/tts_transformer-en-200_speaker-cv4",
arg_overrides={"vocoder": "griffin_lim", "fp16": False}
)

I believe this is due to a missing hifigan.bin / hifigan.json checkpoint in the repos. Other similar repos don't throw this error, such as https://huggingface.co/facebook/fastspeech2-en-200_speaker-cv4/tree/main

So either the documentation should change or a hifigan checkpoint should be added. Adding a hifigan is likely preferable because hifigan's do seem to sound much better than griffin_lim.