Universal Semantic Disentangled Privacy-preserving Speech Representation Learning
This addresses privacy concerns for users and developers in speech-based AI applications, though it appears incremental as it builds on existing codec methods with a focus on disentanglement and evaluation.
The authors tackled the privacy risks of using speech data to train large language models by proposing the Universal Speech Codec (USC), which disentangles speech into privacy-preserving semantic representations and residual acoustic ones, achieving state-of-the-art reconstruction while removing identifiable speaker attributes.
The use of audio recordings of human speech to train LLMs poses privacy concerns due to these models' potential to generate outputs that closely resemble artifacts in the training data. In this study, we propose a speaker privacy-preserving representation learning method through the Universal Speech Codec (USC), a computationally efficient encoder-decoder model that disentangles speech into: (i) privacy-preserving semantically rich representations, capturing content and speech paralinguistics, and (ii) residual acoustic and speaker representations that enables high-fidelity reconstruction. Extensive evaluations presented show that USC's semantic representation preserves content, prosody, and sentiment, while removing potentially identifiable speaker attributes. Combining both representations, USC achieves state-of-the-art speech reconstruction. Additionally, we introduce an evaluation methodology for measuring privacy-preserving properties, aligning with perceptual tests. We compare USC against other codecs in the literature and demonstrate its effectiveness on privacy-preserving representation learning, illustrating the trade-offs of speaker anonymization, paralinguistics retention and content preservation in the learned semantic representations. Audio samples are shared in https://www.amazon.science/usc-samples.