It has been shown that the intelligibility of noisy speech can be improved by speech enhancement (SE) algorithms. However, monaural SE has not been established as an effective frontend for automatic speech recognition (ASR) in noisy conditions compared to an ASR model trained on noisy speech directly. The divide between SE and ASR impedes the progress of robust ASR systems, especially as SE has made major advances in recent years. This paper focuses on eliminating this divide with an ARN (attentive recurrent network) time-domain and a CrossNet time-frequency domain enhancement models. The proposed systems fully decouple frontend enhancement and backend ASR trained only on clean speech. Results on the WSJ, CHiME-2, LibriSpeech, and CHiME-4 corpora demonstrate that ARN and CrossNet enhanced speech both translate to improved ASR results in noisy and reverberant environments, and generalize well to real acoustic scenarios. The proposed system outperforms the baselines trained on corrupted speech directly. Furthermore, it cuts the previous best word error rate (WER) on CHiME-2 by $28.4\%$ relatively with a $5.57\%$ WER, and achieves $3.32/4.44\%$ WER on single-channel CHiME-4 simulated/real test data without training on CHiME-4.
It has been shown that the intelligibility of noisy speech can be improved by speech enhancement algorithms. However, speech enhancement has not been established as an effective front-end for robust automatic speech recognition (ASR) in comparison with an ASR model trained on noisy speech directly. The divide between speech enhancement and ASR impedes the progress of robust ASR systems especially as speech enhancement has made big strides in recent years. In this work, we focus on eliminating such divide with an ARN (attentive recurrent network) based time-domain enhancement model. The proposed system fully decouples speech enhancement and an acoustic model trained only on clean speech. Results on the CHiME-2 corpus show that ARN enhanced speech translates to improved ASR results. The proposed system achieves $6.28\%$ average word error rate, outperforming the previous best by $19.3\%$.
This study addresses robust automatic speech recognition (ASR) by introducing a Conformer-based acoustic model. The proposed model builds on a state-of-the-art recognition system using a bi-directional long short-term memory (BLSTM) model with utterance-wise dropout and iterative speaker adaptation, but employs a Conformer encoder instead of the BLSTM network. The Conformer encoder uses a convolution-augmented attention mechanism for acoustic modeling. The proposed system is evaluated on the monaural ASR task of the CHiME-4 corpus. Coupled with utterance-wise normalization and speaker adaptation, our model achieves $6.25\%$ word error rate, which outperforms the previous best system by $8.4\%$ relatively. In addition, the proposed Conformer-based model is $18.3\%$ smaller in model size and reduces total training time by $79.6\%$.
Large datasets are very useful for training speaker recognition systems, and various research groups have constructed several over the years. Voxceleb is a large dataset for speaker recognition that is extracted from Youtube videos. This paper presents an audio-visual method for acquiring audio data from Youtube given the speaker's name as input. The system follows a pipeline similar to that of the Voxceleb data acquisition method. However, our work focuses on fast data acquisition by using face-tracking in subsequent frames once a face has been detected -- this is preferable over face detection for every frame considering its computational cost. We show that applying audio diarization to our data after acquiring it can yield equal error rates comparable to Voxceleb. A secondary set of experiments showed that we could further decrease the error rate by fine-tuning a pre-trained x-vector system with the acquired data. Like Voxceleb, the work here focuses primarily on developing audio for celebrities. However, unlike Voxceleb, our target audio data is from celebrities in East Asian countries. Finally, we set up a speaker verification task to evaluate the accuracy of our acquired data. After diarization and fine-tuning, we achieved an equal error rate of approximately 4\% across our entire dataset.