When labeled data is insufficient, semi-supervised learning with the pseudo-labeling technique can significantly improve the performance of automatic speech recognition. However, pseudo-labels are often noisy, containing numerous incorrect tokens. Taking noisy labels as ground-truth in the loss function results in suboptimal performance. Previous works attempted to mitigate this issue by either filtering out the nosiest pseudo-labels or improving the overall quality of pseudo-labels. While these methods are effective to some extent, it is unrealistic to entirely eliminate incorrect tokens in pseudo-labels. In this work, we propose a novel framework named alternative pseudo-labeling to tackle the issue of noisy pseudo-labels from the perspective of the training objective. The framework comprises several components. Firstly, a generalized CTC loss function is introduced to handle noisy pseudo-labels by accepting alternative tokens in the positions of incorrect tokens. Applying this loss function in pseudo-labeling requires detecting incorrect tokens in the predicted pseudo-labels. In this work, we adopt a confidence-based error detection method that identifies the incorrect tokens by comparing their confidence scores with a given threshold, thus necessitating the confidence score to be discriminative. Hence, the second proposed technique is the contrastive CTC loss function that widens the confidence gap between the correctly and incorrectly predicted tokens, thereby improving the error detection ability. Additionally, obtaining satisfactory performance with confidence-based error detection typically requires extensive threshold tuning. Instead, we propose an automatic thresholding method that uses labeled data as a proxy for determining the threshold, thus saving the pain of manual tuning.
Recently, there has been increasing progress in end-to-end automatic speech recognition (ASR) architecture, which transcribes speech to text without any pre-trained alignments. One popular end-to-end approach is the hybrid Connectionist Temporal Classification (CTC) and attention (CTC/attention) based ASR architecture. However, how to deploy hybrid CTC/attention systems for online speech recognition is still a non-trivial problem. This article describes our proposed online hybrid CTC/attention end-to-end ASR architecture, which replaces all the offline components of conventional CTC/attention ASR architecture with their corresponding streaming components. Firstly, we propose stable monotonic chunk-wise attention (sMoChA) to stream the conventional global attention, and further propose monotonic truncated attention (MTA) to simplify sMoChA and solve the training-and-decoding mismatch problem of sMoChA. Secondly, we propose truncated CTC (T-CTC) prefix score to stream CTC prefix score calculation. Thirdly, we design dynamic waiting joint decoding (DWJD) algorithm to dynamically collect the predictions of CTC and attention in an online manner. Finally, we use latency-controlled bidirectional long short-term memory (LC-BLSTM) to stream the widely-used offline bidirectional encoder network. Experiments with LibriSpeech English and HKUST Mandarin tasks demonstrate that, compared with the offline CTC/attention model, our proposed online CTC/attention model improves the real time factor in human-computer interaction services and maintains its performance with moderate degradation. To the best of our knowledge, this is the first work to provide the full-stack online solution for CTC/attention end-to-end ASR architecture.
Utilizing the large-scale unlabeled data from the target domain via pseudo-label clustering algorithms is an important approach for addressing domain adaptation problems in speaker verification tasks. In this paper, we propose a novel progressive subgraph clustering algorithm based on multi-model voting and double-Gaussian based assessment (PGMVG clustering). To fully exploit the relationships among utterances and the complementarity among multiple models, our method constructs multiple k-nearest neighbors graphs based on diverse models and generates high-confidence edges using a voting mechanism. Further, to maximize the intra-class diversity, the connected subgraph is utilized to obtain the initial pseudo-labels. Finally, to prevent disastrous clustering results, we adopt an iterative approach that progressively increases k and employs a double-Gaussian based assessment algorithm to decide whether merging sub-classes.
This report describes our submission to track1 and track3 for VoxCeleb Speaker Recognition Challenge 2022(VoxSRC2022). Our best system achieves minDCF 0.1397 and EER 2.414 in track1, minDCF 0.388 and EER 7.030 in track3.
Previous research in speech enhancement has mostly focused on modeling time or time-frequency domain information alone, with little consideration given to the potential benefits of simultaneously modeling both domains. Since these domains contain complementary information, combining them may improve the performance of the model. In this letter, we propose a new approach to simultaneously model time and time-frequency domain information in a single model. We begin with the DPT-FSNet (causal version) model as a baseline and modify the encoder structure by replacing the original encoder with three separate encoders, each dedicated to modeling time-domain, real-imaginary, and magnitude information, respectively. Additionally, we introduce a feature fusion module both before and after the dual-path processing blocks to better leverage information from the different domains. The outcomes of our experiments reveal that the proposed approach achieves superior performance compared to existing state-of-the-art causal models, while preserving a relatively compact model size and low computational complexity.
ECAPA-TDNN is currently the most popular TDNN-series model for speaker verification, which refreshed the state-of-the-art(SOTA) performance of TDNN models. However, one-dimensional convolution has a global receptive field over the feature channel. It destroys the time-frequency relevance of the spectrogram. Besides, as ECAPA-TDNN only has five layers, a much shallower structure compared to ResNet restricts the capability to generate deep representations. To further improve ECAPA-TDNN, we propose a progressive channel fusion strategy that splits the spectrogram across the feature channel and gradually expands the receptive field through the network. Secondly, we enlarge the model by extending the depth and adding branches. Our proposed model achieves EER with 0.718 and minDCF(0.01) with 0.0858 on vox1o, relatively improved 16.1\% and 19.5\% compared with ECAPA-TDNN-large.
Selecting application scenarios matching data is important for the automatic speech recognition (ASR) training, but it is difficult to measure the matching degree of the training corpus. This study proposes a unsupervised target-aware data selection method based on speech corpora divergence (SCD), which can measure the similarity between two speech corpora. We first use the self-supervised Hubert model to discretize the speech corpora into label sequence and calculate the N-gram probability distribution. Then we calculate the Kullback-Leibler divergence between the N-grams as the SCD. Finally, we can choose the subset which has minimum SCD to the target corpus for annotation and training. Compared to previous data selection method, the SCD data selection method can focus on more acoustic details and guarantee the diversity of the selected set. We evaluate our method on different accents from Common Voice. Experiments show that the proposed SCD data selection can realize 14.8% relative improvements to the random selection, comparable or even superior to the result of supervised selection.
Recently, convolutional neural networks (CNNs) have been widely used in sound event detection (SED). However, traditional convolution is deficient in learning time-frequency domain representation of different sound events. To address this issue, we propose multi-dimensional frequency dynamic convolution (MFDConv), a new design that endows convolutional kernels with frequency-adaptive dynamic properties along multiple dimensions. MFDConv utilizes a novel multi-dimensional attention mechanism with a parallel strategy to learn complementary frequency-adaptive attentions, which substantially strengthen the feature extraction ability of convolutional kernels. Moreover, in order to promote the performance of mean teacher, we propose the confident mean teacher to increase the accuracy of pseudo-labels from the teacher and train the student with high confidence labels. Experimental results show that the proposed methods achieve 0.470 and 0.692 of PSDS1 and PSDS2 on the DESED real validation dataset.
In this paper, we propose a two-stage heterogeneous lightweight network for monaural speech enhancement. Specifically, we design a novel two-stage framework consisting of a coarse-grained full-band mask estimation stage and a fine-grained low-frequency refinement stage. Instead of using a hand-designed real-valued filter, we use a novel learnable complex-valued rectangular bandwidth (LCRB) filter bank as an extractor of compact features. Furthermore, considering the respective characteristics of the proposed two-stage task, we used a heterogeneous structure, i.e., a U-shaped subnetwork as the backbone of CoarseNet and a single-scale subnetwork as the backbone of FineNet. We conducted experiments on the VoiceBank + DEMAND and DNS datasets to evaluate the proposed approach. The experimental results show that the proposed method outperforms the current state-of-the-art methods, while maintaining relatively small model size and low computational complexity.
Code-switching automatic speech recognition becomes one of the most challenging and the most valuable scenarios of automatic speech recognition, due to the code-switching phenomenon between multilingual language and the frequent occurrence of code-switching phenomenon in daily life. The ISCSLP 2022 Chinese-English Code-Switching Automatic Speech Recognition (CSASR) Challenge aims to promote the development of code-switching automatic speech recognition. The ISCSLP 2022 CSASR challenge provided two training sets, TAL_CSASR corpus and MagicData-RAMC corpus, a development and a test set for participants, which are used for CSASR model training and evaluation. Along with the challenge, we also provide the baseline system performance for reference. As a result, more than 40 teams participated in this challenge, and the winner team achieved 16.70% Mixture Error Rate (MER) performance on the test set and has achieved 9.8% MER absolute improvement compared with the baseline system. In this paper, we will describe the datasets, the associated baselines system and the requirements, and summarize the CSASR challenge results and major techniques and tricks used in the submitted systems.