Skip to content

Support on-device voice cloning with OmniVoice #1135

@IgorSwat

Description

@IgorSwat

Problem Description

OmniVoice is a highly capable, open source voice cloning model. Tested on various english/polish samples, the model seems to produce a high-quality speech, very similar to the original.

The model weighs over 2 GB, but with a proper quantization it should be feasible to use it on-device within the ExecuTorch runtime.

What should be done

  • Export the model to the ExecuTorch format and optimize for on-device inference (gpu/npu, quantization).
  • Implement voice cloning module in RNE

Benefits to React Native ExecuTorch

A voice cloning module would open the door to building various catchy apps, built around this feature.

Metadata

Metadata

Assignees

Labels

ideaNew idea to enhance the library, suggestion, etc.modelIssues related to exporting, improving, fixing ML models

Projects

No projects

Relationships

None yet

Development

No branches or pull requests

Issue actions