1. Workflow Overview

This workflow converts text to natural speech using Index-TTS, supporting voice cloning and audio enhancement. Key features:

• Text-to-Speech: Processes long texts (e.g., novels) into fluent speech.

• Voice Cloning: Mimics speaker timbre from reference audio (e.g., 蔡徐坤.wav).

• Noise Reduction: Cleans background noise for professional output.

Core Models:

• Index-TTS: Main model for speech synthesis (requires plugin ComfyUI-Index-TTS).

• Audio Tools: Noise removal (AudioCleanupNode), timbre loading (TimbreAudioLoader).

2. Key Nodes & Installation

Dependencies:

• Index-TTS models (~2-3GB) auto-download on first use.

3. Workflow Structure

Group 1: Input & Voice Cloning

• Inputs:

  • Reference audio via LoadAudio.

  • Timbre template via TimbreAudioLoader.

• Steps:

  1. IndexTTSNode generates speech from text (e.g., novel chapters).

  2. Parameters: Speed (1.0), emotion (0.8), seed (1155511506).

Group 2: Post-Processing

• Input: Raw generated audio.

• Steps:

  1. AudioCleanupNode applies noise reduction (100-8000Hz range).

  2. SaveAudio exports WAV to audio/ComfyUI.

Group 3: Preview & Output

• Preview: Listen via PreviewAudio.

• Output: WAV file (e.g., ComfyUI_20240513_142301.wav).

4. Inputs & Outputs

Inputs:

• Text: Supports long texts (example: 4-chapter novel).

• Reference Audio: Clear voice sample (≥10 sec recommended).

• Timbre Template (optional): Style template (e.g., 抖音-读小说.wav).

Output:

• WAV file saved in ComfyUI/audio.

5. Notes

1. VRAM:

   • Index-TTS requires ~4GB VRAM; split long texts if needed.

2. Quality Tips:

   • Adjust frequency_range in AudioCleanupNode to preserve voice clarity.

3. Voice Control:

   • Change seed in IndexTTSNode for different voice variations.

4. Debugging:

   • Avoid special characters in Chinese text to prevent garbled speech.

   • Pre-clean noisy reference audio for better cloning.