Improving Audio Evaluation via Synchronized Multi-Stem Playback

[lekoOwO]

👋 Hello!

First off, thank you for developing this fantastic project, making sharing UVR5 on web possible!

✨ Feature Suggestion

I wanted to suggest a potential improvement that could greatly enhance the user experience, particularly when evaluating the quality of the separated audio stems.

Currently, when the model outputs multiple tracks (Vocals, Drums, etc.), it's challenging to hear them played back together simultaneously.

The core idea is to enable synchronized playback and real-time volume mixing on the front end.

Why this feature would be helpful:

1. Accurate Evaluation: It would allow users to instantly hear the full reconstructed mix, making it much easier to assess if the separated tracks are perfectly synchronized and free of artifacts or phase issues.
2. User Control: Users could adjust the relative volume of each track in real-time. This is extremely valuable for diagnosing signal bleed (leakage) between stems, which is key to model tuning.

Technical Note (for the developer):

Since Gradio doesn't offer this feature natively, this typically requires using a custom solution involving the gr.HTML component to inject a small amount of JavaScript using the Web Audio API. This API is ideally suited for this task as it can handle:

• Precise timing for synchronization.
• Real-time volume control via GainNode without latency.

🙏 Conclusion

I understand this involves custom frontend work, but I believe adding a Multi-Stem Mixer UI would transform the demo from a static output viewer into a much more interactive and professional tool for audio processing.

Thank you again for your hard work!