For efficient transmission of speech signals, speech compression methodologies have attracted significant research attention for decades and are widely used in automatic speech recognition (ASR) services. However, most speech codecs are perception-oriented, leaving redundant information and introducing distortion, which harms ASR systems. Recently, the emergence of neural network-based models has significantly advanced the progress of ASR systems and speech coding, laying the foundation for building a speech compression method specially optimized for ASR systems. In this letter, we propose an ASR-oriented Speech Quantization (ASQ) method to reduce communication costs for speech recognition systems. In the proposed method, a speech quantization model first converts the speech into low bit rate tokens. Then the tokens are transmitted to the server and recognized by a quantized speech recognition model. The two models could be jointly trained in the end-to-end (E2E) style. To mitigate the performance degradation introduced by the quantization components, we design an entropy-guided 3-stage training method that encourages the model to fully utilize the token space and promote recognition accuracy. Experiment results on the LibriSpeech corpus show that compared to an existing non-quantized ASR model with a 256 kbps transmission bit rate, the proposed method can achieve a transmission bit rate of 0.6 kbps without any influence on word error rate (WER). It also significantly surpasses the 2-step pipeline that first performs speech codec and then recognizes with a several times lower bit rate.

Serialized Output Training (SOT) has emerged as the mainstream approach for addressing the multi-talker overlapped speech recognition challenge due to its simplicity. However, SOT encounters cross-domain performance degradation which hinders its application. Meanwhile, traditional domain adaption methods may harm the accuracy of speaker change point prediction evaluated by UD-CER, which is an important metric in SOT. To solve these issues, we propose Pseudo-Labeling based SOT (PL-SOT) for domain adaptation by treating speaker change token ( $< $ sc $>$ ) specially during training to increase the accuracy of speaker change point prediction. Firstly, we improve CTC loss by proposing Weakening and Enhancing CTC (WE-CTC) loss to weaken the learning of error-prone labels surrounding $< $ sc $>$ while enhance the emission probability of $< $ sc $>$ through modifying posteriors of the pseudo-labels. Secondly, we introduce Weighted Confidence Filter (WCF) that assigns higher scores of $< $ sc $>$ to exclude low-quality pseudo-labels without hurting the $< $ sc $>$ prediction. Experimental results show that PL-SOT achieves 17.7%/12.8% average relative reduction of CER/UD-CER, with AliMeeting as source domain and AISHELL-4 along with MagicData-RAMC as target domain.

Multi-scale context information, especially long-range dependency, has shown to be beneficial for speaker verification (SV) tasks. In this paper, we propose three methods to systematically explore long-range context SV feature extraction based on ResNet and analyze their complementarity. Firstly, the Hierarchical-split block (HS-block) is introduced to enlarge the receptive fields (RFs) and extract long-range context information over the feature maps of a single layer, where the multi-channel feature maps are split into multiple groups and then stacked together. Then, by analyzing the contribution of each location of the convolution kernel to SV, we find the traditional convolution with a square kernel is not effective for long-range feature extraction. Therefore, we propose cross convolution kernel (cross-conv), which replaces the original 3 × 3 convolution kernel with a 1 × 5 and 5 × 1 convolution kernel. Cross-conv further enlarges the RFs with the same FLOPs and parameters. Finally, the Depthwise Separable Self-Attention (DSSA) module uses an explicit sparse attention strategy to capture effective long-range dependencies globally in each channel. Experiments are conducted on the VoxCeleb and CnCeleb to verify the effectiveness and robustness of the proposed system. Experimental results show that the combination of HS-block, cross-conv, and DSSA module achieves better performance than any single method, which demonstrates the complementarity of these three methods.

Most existing direction-aware speech separation systems lead to performance degradation when the angle difference between speakers is small due to the low spatial discrimination. To address this issue, we propose a two-step soft-direction-aware speech separation (So-DAS) framework, which consists of a direction of arrival (DOA) estimation module and a speech separation module. First, the two modules are individually optimized, and directional features (DFs) derived from ground-truth DOAs are utilized as spatial information to facilitate the separation module. Next, the two modules are cascaded and optimized with only separation loss, and the DFs are generated using the estimator outputs. By this means, the consistency between the two modules is strengthened, and thus spatial cues that are more beneficial to the separation task can be exploited by the network itself. The experimental results show that compared to the baselines, DFs extracted by our proposed method provides clearer superiority, especially when the angle difference between speakers is small. In addition, our approach yields a state-of-the-art word error rate of 3.4% on the real-recorded utterance-wise LibriCSS dataset.

This paper proposes a new quantization framework on activation function of deep neural networks (DNN). We implement fixed-point DNN by quantizing the activations into powers-of-two integers. The costly multiplication operations in using DNN can be replaced with low-cost bit-shifts to massively save computations. Thus, applying DNN-based speech recognition on embedded systems becomes much easier. Experiments show that the proposed method leads to no performance degradation.

Learning to speak a foreign language is not an easy task for many people. This paper describes approaches to automatic objective assessment of pronunciation quality. The approaches described here can be classified into two categories, text-dependent and text-independent, according to whether a teacher's voice is present. In the text-independent one, algorithms based on energy and pitch contour are introduced. Also, the average rate of variation in energy and pitch frequency, mean subtracted energy and pitch frequency are used as main features. Compared to the previously reported approach using average phone segment posterior probabilities, the new approach achieves favorable performance on the same test set.