[Feature request] Support for Quantized ONNX Model Conversion for Stream Inference

[TranDacKhoa]

🚀 Feature Description

Is there support in Coqui TTS for converting models to a quantized ONNX format for stream inference? This feature would enhance model performance and reduce inference time for real-time applications.

Solution

Implement a workflow or tool within Coqui TTS for easy conversion of TTS models to quantized ONNX format.

Alternative Solutions

Currently, external tools like ONNX Runtime or TensorRT can be used for post-conversion quantization, but having this feature natively would streamline the process.

Additional context

Any existing documentation or insights on this topic would be appreciated. Thank you!