Compute-efficient Real-time Voice Cloning

The project is to significantly reduce the computation cost of machine-learning based real-time voice cloning. This can be achieved by applying both software and hardware-aware techniques such as quantization, pruning, and knowledge distillation. This enables voice cloning to be deployed in mobile and embedded systems.

Objectives

  • Improved voice cloning software with reduced computation cost
  • Demonstration on laptop/desktop
  • (Optional) Demonstration on mobile devices

Motivations

Real-time voice cloning can be extremely useful in many applications, such as providing more personalized voice assistance as well as helping people who have lost their voices to synthesize speech from text that reassembles their voices. Current voice-cloning frameworks, however, involve heavy computation that may need powerful processors to run, thus limiting their usefulness.

Qualifications

Minimum Qualifications:
None Listed

Preferred Qualifications:
None Listed


Details

Project Partner:

Lizhong Chen

NDA/IPA:

No Agreement Required

Number Groups:

1

Project Status:

Accepting Applicants

Card Image Capstone