The expectation to deploy a universal neural network for speech enhancement, with the aim of improving noise robustness across diverse speech processing tasks, faces challenges due to the existing lack of awareness within static speech enhancement frameworks regarding the expected speech in downstream modules. These limitations impede the effectiveness of static speech enhancement approaches in achieving optimal performance for a range of speech processing tasks, thereby challenging the notion of universal applicability. The fundamental issue in achieving universal speech enhancement lies in effectively informing the speech enhancement module about the features of downstream modules. In this study, we present a novel weighting prediction approach, which explicitly learns the task relationships from downstream training information to address the core challenge of universal speech enhancement. We found the role of deciding whether to employ data augmentation techniques as crucial downstream training information. This decision significantly impacts the expected speech and the performance of the speech enhancement module. Moreover, we introduce a novel speech enhancement network, the Plugin Speech Enhancement (Plugin-SE). The Plugin-SE is a dynamic neural network that includes the speech enhancement module, gate module, and weight prediction module. Experimental results demonstrate that the proposed Plugin-SE approach is competitive or superior to other joint training methods across various downstream tasks.
Cascading multiple pre-trained models is an effective way to compose an end-to-end system. However, fine-tuning the full cascaded model is parameter and memory inefficient and our observations reveal that only applying adapter modules on cascaded model can not achieve considerable performance as fine-tuning. We propose an automatic and effective adaptive learning method to optimize end-to-end cascaded multi-task models based on Neural Architecture Search (NAS) framework. The candidate adaptive operations on each specific module consist of frozen, inserting an adapter and fine-tuning. We further add a penalty item on the loss to limit the learned structure which takes the amount of trainable parameters into account. The penalty item successfully restrict the searched architecture and the proposed approach is able to search similar tuning scheme with hand-craft, compressing the optimizing parameters to 8.7% corresponding to full fine-tuning on SLURP with an even better performance.
Self-supervised pre-trained models such as HuBERT and WavLM leverage unlabeled speech data for representation learning and offer significantly improve for numerous downstream tasks. Despite the success of these methods, their large memory and strong computational requirements hinder their application on resource restricted devices. Therefore, this paper introduces GenDistiller, a novel knowledge distillation framework to distill hidden representations from teacher network based on generative language model. The generative structure enables the proposed model to generate the target teacher hidden layers autoregressively, considering the interactions between hidden layers without instroducing additional inputs. A two-dimensional attention mechanism is implemented to ensure the causality of hidden layers, while preserving bidirectional attention in the time dimension. Experiments reveal the advantage of the generative distiller over the baseline system that predicts the hidden layers of teacher network directly without a generatvie model.
The performance of the keyword spotting (KWS) system based on audio modality, commonly measured in false alarms and false rejects, degrades significantly under the far field and noisy conditions. Therefore, audio-visual keyword spotting, which leverages complementary relationships over multiple modalities, has recently gained much attention. However, current studies mainly focus on combining the exclusively learned representations of different modalities, instead of exploring the modal relationships during each respective modeling. In this paper, we propose a novel visual modality enhanced end-to-end KWS framework (VE-KWS), which fuses audio and visual modalities from two aspects. The first one is utilizing the speaker location information obtained from the lip region in videos to assist the training of multi-channel audio beamformer. By involving the beamformer as an audio enhancement module, the acoustic distortions, caused by the far field or noisy environments, could be significantly suppressed. The other one is conducting cross-attention between different modalities to capture the inter-modal relationships and help the representation learning of each modality. Experiments on the MSIP challenge corpus show that our proposed model achieves 2.79% false rejection rate and 2.95% false alarm rate on the Eval set, resulting in a new SOTA performance compared with the top-ranking systems in the ICASSP2022 MISP challenge.
Spoken language understanding (SLU) treats automatic speech recognition (ASR) and natural language understanding (NLU) as a unified task and usually suffers from data scarcity. We exploit an ASR and NLU joint training method based on meta auxiliary learning to improve the performance of low-resource SLU task by only taking advantage of abundant manual transcriptions of speech data. One obvious advantage of such method is that it provides a flexible framework to implement a low-resource SLU training task without requiring access to any further semantic annotations. In particular, a NLU model is taken as label generation network to predict intent and slot tags from texts; a multi-task network trains ASR task and SLU task synchronously from speech; and the predictions of label generation network are delivered to the multi-task network as semantic targets. The efficiency of the proposed algorithm is demonstrated with experiments on the public CATSLU dataset, which produces more suitable ASR hypotheses for the downstream NLU task.
Siamese networks have shown effective results in unsupervised visual representation learning. These models are designed to learn an invariant representation of two augmentations for one input by maximizing their similarity. In this paper, we propose an effective Siamese network to improve the robustness of End-to-End automatic speech recognition (ASR). We introduce spatial-temporal dropout to support a more violent disturbance for Siamese-ASR framework. Besides, we also relax the similarity regularization to maximize the similarities of distributions on the frames that connectionist temporal classification (CTC) spikes occur rather than on all of them. The efficiency of the proposed architecture is evaluated on two benchmarks, AISHELL-1 and Librispeech, resulting in 7.13% and 6.59% relative character error rate (CER) and word error rate (WER) reductions respectively. Analysis shows that our proposed approach brings a better uniformity for the trained model and enlarges the CTC spikes obviously.
Joint training of speech enhancement model (SE) and speech recognition model (ASR) is a common solution for robust ASR in noisy environments. SE focuses on improving the auditory quality of speech, but the enhanced feature distribution is changed, which is uncertain and detrimental to the ASR. To tackle this challenge, an approach with multiple confidence gates for jointly training of SE and ASR is proposed. A speech confidence gates prediction module is designed to replace the former SE module in joint training. The noisy speech is filtered by gates to obtain features that are easier to be fitting by the ASR network. The experimental results show that the proposed method has better performance than the traditional robust speech recognition system on test sets of clean speech, synthesized noisy speech, and real noisy speech.
The harmonic structure of speech is resistant to noise, but the harmonics may still be partially masked by noise. Therefore, we previously proposed a harmonic gated compensation network (HGCN) to predict the full harmonic locations based on the unmasked harmonics and process the result of a coarse enhancement module to recover the masked harmonics. In addition, the auditory loudness loss function is used to train the network. For the DNS Challenge, we update HGCN with the following aspects, resulting in HGCN+. First, a high-band module is employed to help the model handle full-band signals. Second, cosine is used to model the harmonic structure more accurately. Then, the dual-path encoder and dual-path rnn (DPRNN) are introduced to take full advantage of the features. Finally, a gated residual linear structure replaces the gated convolution in the compensation module to increase the receptive field of frequency. The experimental results show that each updated module brings performance improvement to the model. HGCN+ also outperforms the referenced models on both wide-band and full-band test sets.
Mask processing in the time-frequency (T-F) domain through the neural network has been one of the mainstreams for single-channel speech enhancement. However, it is hard for most models to handle the situation when harmonics are partially masked by noise. To tackle this challenge, we propose a harmonic gated compensation network (HGCN). We design a high-resolution harmonic integral spectrum to improve the accuracy of harmonic locations prediction. Then we add voice activity detection (VAD) and voiced region detection (VRD) to the convolutional recurrent network (CRN) to filter harmonic locations. Finally, the harmonic gating mechanism is used to guide the compensation model to adjust the coarse results from CRN to obtain the refinedly enhanced results. Our experiments show HGCN achieves substantial gain over a number of advanced approaches in the community.