This repository provides a comprehensive evaluation of voice cloning models based on objective speech quality metrics. Our goal is to assess the effectiveness of these models in generating high-quality, intelligible, and natural-sounding voices.
| Model | PESQ | STOI | MCD | Pitch Corr | Spec Conv | Energy Ratio | SNR (dB) |
|---|---|---|---|---|---|---|---|
| OpenVoice | 1.165 | 0.136 | 37.988 | -0.027 | 3.475 | 12.305 | -11.193 |
| CoquiTTS | 1.727 | 0.143 | 203.193 | 0.012 | 6.675 | 45.896 | -16.717 |
| F5-TTS | 1.782 | 0.171 | 174.265 | 0.060 | 6.082 | 39.209 | -16.065 |
| E2-TTS | 2.281 | 0.165 | 158.578 | -0.051 | 5.760 | 34.939 | -15.551 |
- PESQ (Perceptual Evaluation of Speech Quality): Measures speech quality, with values ranging from -0.5 to 4.5 (higher is better).
- STOI (Short-Time Objective Intelligibility): Assesses how well the synthesized voice is understood (range: 0 to 1, higher is better).
- MCD (Mel Cepstral Distortion): Lower values indicate more accurate voice cloning.
- Pitch Correlation: Measures how closely the pitch matches the original speaker (closer to 1 is better).
- Spectral Convergence (Spec Conv): Evaluates how well spectral features align (lower is better).
- Energy Ratio: Assesses energy distribution in frequency bands.
- SNR (Signal-to-Noise Ratio in dB): Higher values indicate cleaner, more natural output.
You can experiment with each model using the provided links:
- Best for natural voice quality: E2-TTS (highest PESQ, lowest MCD).
- Best for intelligibility: F5-TTS (highest STOI score).
- Moderate performance: CoquiTTS (balanced results but high spectral distortion).
- Least recommended: OpenVoice (low PESQ, less realistic output).
This project is licensed under the Apache 2.0 - see the LICENSE file for details.