Codec2Vec: Unveiling Speech Representations with Neural Codecs

This research introduces a novel self-supervised approach to speech representation learning, leveraging neural speech codecs. The approach is likely to improve downstream speech tasks by providing richer and more robust representations of audio data.