search

NVIDIA / radtts

Provides training, inference and voice conversion recipes for RADTTS and RADTTS++: Flow-based TTS models with Robust Alignment Learning, Diverse Synthesis, and Generative Modeling and Fine-Grained Control over of Low Dimensional (F0 and Energy) Speech Attributes.
MIT License
280 stars 40 forks source link

Trouble with inferencing without pitch and energy condition #14

Open tenebo opened 1 year ago

tenebo commented 1 year ago

Hi, I trained LJSpeech datasets without pitch and energy conditioning, and tried to inference it. However, I got AttributeError: 'RADTTS' object has no attribute 'dur_pred_layer' error.

When training, I followed the code as stated in the readme. And here is the colab notebook that I tried to inference: https://colab.research.google.com/drive/1pwqjZri7k_hoLUNFK3Cxmk12BPad67Kb?usp=sharing

Thank You

tenebo commented 1 year ago

It seems like error occurs as get_attribute_prediction_model does not support inferencing without pitch and energy conditioning. I hope the bug is fixed soon.

https://github.com/NVIDIA/radtts/blob/main/radtts.py#L188-L189 https://github.com/NVIDIA/radtts/blob/07759cd474458f46db45cab975a85ba21b7fee0a/attribute_prediction_model.py#L30

HobisPL commented 1 year ago

You need use config predictor dap/agab/bgap configljs{agap,bgap,dap}.json