[HELP] Audio evaluation and model quantization!

[Dong-mel]

I am a beginner in TTS and have some questions when using Coqui-TTS. Hope somebody can help me.

1. How to evaluate the quality of synthesized audio?
I found a GitHub project that contains some evaluation methods, but I don’t know if this is accurate.

2. TTS model quantization
I tried fine-tuning a model and found that the resulting model size was too large (800+ MB, model type: vits), but the model downloaded from hugging Face was around 100MB, so I wanted to try to reduce its size.
Searching the forums, I found that someone had asked the same question discussions#2991, but it seems that there is no good result.

Does anyone have some documentation or experience to share?