Unsupervised Neural Mask Estimator For Generalized Eigen-Value Beamforming Based ASR
This work addresses the challenge of reducing reliance on supervised data for training beamforming models in ASR, which is incremental but beneficial for real-world noisy speech applications.
The paper tackles the problem of training neural mask estimators for acoustic beamforming in multi-channel ASR without requiring supervised clean recordings, by using models based on signal enhancement and beamforming as mask estimates. The approach achieves ASR performances significantly better than a teacher model trained on an out-of-domain dataset and on par with oracle mask estimators trained on in-domain data, as demonstrated on the CHiME-3 and REVERB challenge corpora.
The state-of-art methods for acoustic beamforming in multi-channel ASR are based on a neural mask estimator that predicts the presence of speech and noise. These models are trained using a paired corpus of clean and noisy recordings (teacher model). In this paper, we attempt to move away from the requirements of having supervised clean recordings for training the mask estimator. The models based on signal enhancement and beamforming using multi-channel linear prediction serve as the required mask estimate. In this way, the model training can also be carried out on real recordings of noisy speech rather than simulated ones alone done in a typical teacher model. Several experiments performed on noisy and reverberant environments in the CHiME-3 corpus as well as the REVERB challenge corpus highlight the effectiveness of the proposed approach. The ASR results for the proposed approach provide performances that are significantly better than a teacher model trained on an out-of-domain dataset and on par with the oracle mask estimators trained on the in-domain dataset.