English | 简体中文

ComfyUI_FishSpeech_EX

This plugin is optimized for Fish-Speech-1.5 version and is only applicable to version 1.5:

1. The plugin references the ComfyUI-fish-speech plugin for optimization, and modifies the overall configuration address and installation method.
2. The required Python libraries for the plugin have been improved, mainly vector-quantize-pytorch. If this library is not installed, the audio quality will be poor. This problem has been bothering me for a few days, and I searched the entire FishSpeech plugin to find the sampling step issue. If this problem has also been bothering you, please give it a like, thank you!

Specific nodes：

• EX_AudioToPrompt

1. audio: ComfyUI audio.
2. vqgan: VQGAN model.
3. restored_audio: Decoded audio.
4. prompt_tokens: Tokens corresponding to the prompt audio.

• EX_Prompt2Semantic

1. prompt_tokens: The token corresponding to the input prompt audio.
2. codes: The generated audio Code.

• EX_LoadVQGAN Load the VQGAN model, input the model path, and output the model.

• EX_Semantic2Image Analyze audio Code, output corresponding audio.

• EX_SaveAudioToMp3 Save the audio to an MP3 file.

Work flow

Reference materials

• AnyaCoder/ComfyUI-fish-speech - Official Implementaion
• fishaudio/fish-speech - SOTA Open Source TTS.