encoder2teach | language models , such as BERT , to teach Wav2vec2

encoder2teach is a Python library. encoder2teach has no bugs, it has no vulnerabilities, it has build file available and it has low support. You can download it from GitHub.

We explore the potential for language models, such as BERT, to teach Wav2vec2 representation learning. Creating audio data for automatic speech recognition (ASR) can be challenging, especially when large quantities of data are needed. While textual data is easier to collect, language models have demonstrated impressive results in learning contextual representations that are useful for a range of applications. The central question of this study is whether language models can effectively teach models like Wav2vec2 to learn representations. The proposed approach involves freezing a pre-trained language model and comparing its output representation with a student model that will learn to read the representation.