MBZUAI

/

ArTST

like 4

Follow

Mohamed Bin Zayed University of Artificial Intelligence 335

First, thank you for making the weights publicly available!

Request

Would it be possible to include the following configuration files with the model weights:

1. generation_config.json - for consistent generation settings
2. preprocessor_config.json - for proper text preprocessing
3. config.json - for complete model configuration

It would reduce the chance of audio generation issues due to misconfigurations

Currently using:

• CLARTTS_ArTST_TTS.pt
• CLARTTS_ArTSTstar_TTS.pt

Request for Sample Inference Code

Would greatly appreciate:

1. A complete working inference script showing:
   • Correct model loading process
   • Proper speaker embedding handling
   • Text-to-speech generation
   • Audio saving
2. A few example outputs (wav files) generated using this script
3. Expected audio parameters (sample rate, duration, etc.)

Specific Questions

• Which speaker embeddings work best with the model?
• Are there any specific preprocessing steps needed for Arabic text?
• What are the expected audio characteristics of a successful generation?

Having this sample code would help ensure users are implementing the model as intended and getting the expected quality of speech synthesis.

CC @herwoww @amupd