Performing Anonymization

The DPVC library is organized around two classes: Anonymizer, which performs anonymization, and VoiceControlWrapper, a base class for objects that wrap voice control systems. The anonymizer provides an anonymize method to perform anonymization; extensions of VoiceControlWrapper provide inference and extract_embedding methods based on the underlying voice control system.

Example: OpenVoice

DPVC provides a wrapper around the OpenVoice voice control system. The OpenVoiceWrapper class extends VoiceControlWrapper and exposes methods for inference and extraction of speaker embeddings using OpenVoice. A minimal example of using it is as follows:

import dpvc

# Construct the wrapper
vc_wrapper = dpvc.OpenVoiceWrapper()

# Construct the anonymizer
anonymizer = dpvc.Anonymizer(vc_wrapper)

# Perform anonymization
anonymizer.anonymize(src_path, output_path, noise_level=1.0)

Here, src_path should be an input .wav file name, and output_path should be the output .wav file name. The noise_level parameter controls how much noise is added in the differential privacy step. The OpenVoiceDPWrapper object encapsulates the OpenVoice models, and the anonymize method performs the anonymization via differential privacy.