TEVR: Improving Speech Recognition by Token Entropy Variance Reduction
This work addresses speech recognition accuracy for German, with potential applications in privacy-preserving offline virtual assistants, though it appears incremental as it builds on existing language model integration approaches.
The paper tackles speech recognition by reducing token entropy variance relative to the language model, enabling the acoustic model to focus less on tokens the language model can predict reliably. On CommonVoice German, their 900M-parameter TEVR model achieves a 3.64% word error rate, a 16.89% relative reduction compared to previous best results.
This paper presents TEVR, a speech recognition model designed to minimize the variation in token entropy w.r.t. to the language model. This takes advantage of the fact that if the language model will reliably and accurately predict a token anyway, then the acoustic model doesn't need to be accurate in recognizing it. We train German ASR models with 900 million parameters and show that on CommonVoice German, TEVR scores a very competitive 3.64% word error rate, which outperforms the best reported results by a relative 16.89% reduction in word error rate. We hope that releasing our fully trained speech recognition pipeline to the community will lead to privacy-preserving offline virtual assistants in the future.