task: Use VieTTS if output language is Vietnamese

[hahuyhoang411]

Goal

Making the vieTTS for the learning then use this to expand on multilingual TTS

Tasklist

• Chose architecture (MaskGCT, F5, etc)
• Setup training pipeline
• Experiment with model hyperparameters
• Draft research report

[dan-homebrew]

Discussion

• How will we detect outputs in Vietnamese, and switch to VietTTS?
• Future architecture: Ichigo will integrated speech decoder (i.e. built-in TTS)

[tikikun]

Most model that comes with Vietnamese language will have English in it as well. The most realistic solution to this is just having a toggle for multi-lingual capability and use the model that is having Vietnamese language vs just English (English quality will be down a bit but this is more realistic to achieve within this milestone).

I will deep dive more on owning over the TTS part with my f5-tts research.