Feature:Voice Cloning Similarity & Stability Evaluation Report

Is your feature request related to a problem? Please describe.

When testing voice cloning or voice design workflows, it can be difficult to objectively compare generations across prompts, checkpoints, or fine-tuning runs.

Users often rely on subjective listening, making it hard to identify whether changes improve speaker similarity, style consistency, or speech stability.

Describe the solution you'd like

Add an optional evaluation mode that generates a report alongside synthesized audio.

Potential metrics:

• Speaker similarity score
• Speaking rate
• Pitch statistics
• Style consistency across generations
• Word error rate (if reference text is available)

The report could help users compare multiple generations and better evaluate fine-tuning results.

Describe alternatives you've considered

External evaluation tools exist, but they require additional setup and often do not integrate directly with the VoxCPM workflow.

Additional context

This would be especially useful for benchmarking voice cloning, voice design, and future fine-tuning experiments while providing more reproducible evaluations than subjective listening alone.

No files, tests, or entry points are named. Start by tracing the VoxCPM voice cloning and voice design workflow, then determine how an optional evaluation mode could generate reports for synthesized audio and compare generations; done should include agreed metrics and reproducible evaluation output.