Cant push (basically) the same model as TTS
Hey man, I'm trying to finish the audio course and I'm kind of stuck on the third, text to speech exercise. I think it's because when I push models they are labeled as Text to Audio, but you somehow got it working for this model.. It's clearly labeled as Text-to-Speech, so I wanted to ask is there anything you've done specially different with this model? Have you experienced the same? Do you have any advices for me?
Thanks in advance!
I am not clear with your problem. In TTS task I try to use "cmn_hans_cn", but the output is just noise( passed). So I guess I need change cn to pingyin first. As TTS task has no baseline_metric, just do as the https://huggingface.co./learn/audio-course/chapter6/fine-tuning will be ok I think.
kwargs = {
"dataset_tags": "facebook/voxpopuli",
"dataset": "VoxPopuli", # a 'pretty' name for the training dataset
"dataset_args": "config: nl, split: train",
"language": "nl",
"model_name": "SpeechT5 TTS Dutch", # a 'pretty' name for your model
"finetuned_from": "microsoft/speecht5_tts",
"tasks": "text-to-speech",
"tags": "text-to-speech",
}
Thank you! It seems that beside the task text-to-speech you also have to set tags to text-to-speech, otherwise the model will be tagged as Text-To-Audio and it won't go into consideration for the course exercises.
But yeah, the models still output pure noise..