AN EFFICIENT AND ROBUST MULTI-STREAM FRAMEWORK FOR END-TO-END SPEECH RECOGNITION

In voice-enabled domestic or meeting environments, distributed microphone arrays aim to process distant-speech interaction into text with high accuracy. However, with dynamic corruption of noises and reverberations or human movement present, there is no guarantee that any microphone array (stream) is constantly informative. In these cases, an appropriate strategy to dynamically fuse streams is necessary. The multi-stream paradigm in Automatic Speech Recognition (ASR) considers scenarios where parallel streams carry diverse or complementary task-related knowledge. Such streams could be defined as microphone arrays, frequency bands, various modalities or etc. Hence, a robust stream fusion is crucial to emphasize on more informative streams than corrupted ones, especially under unseen conditions. This thesis focuses on improving the performance and robustness of speech recognition in multi-stream scenarios. With increasing use of Deep Neural Networks (DNNs) in ASR, End-to-End (E2E) approaches, which directly transcribe human speech into text, have received greater attention. In this thesis, a multi-stream framework is presented based on the joint Connectionist Temporal Classification/ATTention (CTC/ATT) E2E model, where parallel streams are represented by separate encoders. On top of regular attention networks, a secondary stream-fusion network is to steer the decoder toward the most informative streams. The MEM-Array model aims at improving the far-field ASR robustness using microphone arrays which are activated by separate encoders. With an increasing number of streams (encoders) requiring substantial memory and massive amounts of parallel data, a practical two-stage training strategy is designated to address these issues. Furthermore, a two-stage augmentation scheme is present to improve robustness of the multi-stream model. In MEM-Res, two heterogeneous encoders with different architectures, temporal resolutions and separate CTC networks work in parallel to extract complementary information from the same acoustics. Compared with the best single-stream performance, both models have achieved substantial improvement, outperforming alternative fusion strategies. While the proposed framework optimizes information in multi-stream scenarios, this thesis also studies the Performance Monitoring (PM) measures to predict if recognition results of an E2E model are reliable without growth-truth knowledge. Four PM techniques are investigated, suggesting that PM measures on attention distributions and decoder posteriors are well-correlated with true performances