Qiao Liang

Rajeev Rikhye, Quan Wang, Qiao Liang et al.

In this paper, we propose a solution to allow speaker conditioned speech models, such as VoiceFilter-Lite, to support an arbitrary number of enrolled users in a single pass. This is achieved by using an attention mechanism on multiple speaker embeddings to compute a single attentive embedding, which is then used as a side input to the model. We implemented multi-user VoiceFilter-Lite and evaluated it for three tasks: (1) a streaming automatic speech recognition (ASR) task; (2) a text-independent speaker verification task; and (3) a personalized keyphrase detection task, where ASR has to detect keyphrases from multiple enrolled users in a noisy environment. Our experiments show that, with up to four enrolled users, multi-user VoiceFilter-Lite is able to significantly reduce speech recognition and speaker verification errors when there is overlapping speech, without affecting performance under other acoustic conditions. This attentive speaker embedding approach can also be easily applied to other speaker-conditioned models such as personal VAD and personalized ASR.

Liang Qiao, Zaisheng Li, Zhanzhan Cheng et al.

Table structure recognition is a challenging task due to the various structures and complicated cell spanning relations. Previous methods handled the problem starting from elements in different granularities (rows/columns, text regions), which somehow fell into the issues like lossy heuristic rules or neglect of empty cell division. Based on table structure characteristics, we find that obtaining the aligned bounding boxes of text region can effectively maintain the entire relevant range of different cells. However, the aligned bounding boxes are hard to be accurately predicted due to the visual ambiguities. In this paper, we aim to obtain more reliable aligned bounding boxes by fully utilizing the visual information from both text regions in proposed local features and cell relations in global features. Specifically, we propose the framework of Local and Global Pyramid Mask Alignment, which adopts the soft pyramid mask learning mechanism in both the local and global feature maps. It allows the predicted boundaries of bounding boxes to break through the limitation of original proposals. A pyramid mask re-scoring module is then integrated to compromise the local and global information and refine the predicted boundaries. Finally, we propose a robust table structure recovery pipeline to obtain the final structure, in which we also effectively solve the problems of empty cells locating and division. Experimental results show that the proposed method achieves competitive and even new state-of-the-art performance on several public benchmarks.

Peng Zhang, Can Li, Liang Qiao et al.

Document layout analysis is crucial for understanding document structures. On this task, vision and semantics of documents, and relations between layout components contribute to the understanding process. Though many works have been proposed to exploit the above information, they show unsatisfactory results. NLP-based methods model layout analysis as a sequence labeling task and show insufficient capabilities in layout modeling. CV-based methods model layout analysis as a detection or segmentation task, but bear limitations of inefficient modality fusion and lack of relation modeling between layout components. To address the above limitations, we propose a unified framework VSR for document layout analysis, combining vision, semantics and relations. VSR supports both NLP-based and CV-based methods. Specifically, we first introduce vision through document image and semantics through text embedding maps. Then, modality-specific visual and semantic features are extracted using a two-stream network, which are adaptively fused to make full use of complementary information. Finally, given component candidates, a relation module based on graph neural network is incorported to model relations between components and output final results. On three popular benchmarks, VSR outperforms previous models by large margins. Code will be released soon.

Sankaran Panchapagesan, Daniel S. Park, Chung-Cheng Chiu et al.

Knowledge Distillation is an effective method of transferring knowledge from a large model to a smaller model. Distillation can be viewed as a type of model compression, and has played an important role for on-device ASR applications. In this paper, we develop a distillation method for RNN-Transducer (RNN-T) models, a popular end-to-end neural network architecture for streaming speech recognition. Our proposed distillation loss is simple and efficient, and uses only the "y" and "blank" posterior probabilities from the RNN-T output probability lattice. We study the effectiveness of the proposed approach in improving the accuracy of sparse RNN-T models obtained by gradually pruning a larger uncompressed model, which also serves as the teacher during distillation. With distillation of 60% and 90% sparse multi-domain RNN-T models, we obtain WER reductions of 4.3% and 12.1% respectively, on a noisy FarField eval set. We also present results of experiments on LibriSpeech, where the introduction of the distillation loss yields a 4.8% relative WER reduction on the test-other dataset for a small Conformer model.

Peng Zhang, Yunlu Xu, Zhanzhan Cheng et al.

Since real-world ubiquitous documents (e.g., invoices, tickets, resumes and leaflets) contain rich information, automatic document image understanding has become a hot topic. Most existing works decouple the problem into two separate tasks, (1) text reading for detecting and recognizing texts in images and (2) information extraction for analyzing and extracting key elements from previously extracted plain text. However, they mainly focus on improving information extraction task, while neglecting the fact that text reading and information extraction are mutually correlated. In this paper, we propose a unified end-to-end text reading and information extraction network, where the two tasks can reinforce each other. Specifically, the multimodal visual and textual features of text reading are fused for information extraction and in turn, the semantics in information extraction contribute to the optimization of text reading. On three real-world datasets with diverse document images (from fixed layout to variable layout, from structured text to semi-structured text), our proposed method significantly outperforms the state-of-the-art methods in both efficiency and accuracy.

Zhaofeng Wu, Ding Zhao, Qiao Liang et al.

In automatic speech recognition (ASR), model pruning is a widely adopted technique that reduces model size and latency to deploy neural network models on edge devices with resource constraints. However, multiple models with different sparsity levels usually need to be separately trained and deployed to heterogeneous target hardware with different resource specifications and for applications that have various latency requirements. In this paper, we present Dynamic Sparsity Neural Networks (DSNN) that, once trained, can instantly switch to any predefined sparsity configuration at run-time. We demonstrate the effectiveness and flexibility of DSNN using experiments on internal production datasets with Google Voice Search data, and show that the performance of a DSNN model is on par with that of individually trained single sparsity networks. Our trained DSNN model, therefore, can greatly ease the training process and simplify deployment in diverse scenarios with resource constraints.

Tara N. Sainath, Yanzhang He, Bo Li et al.

Thus far, end-to-end (E2E) models have not been shown to outperform state-of-the-art conventional models with respect to both quality, i.e., word error rate (WER), and latency, i.e., the time the hypothesis is finalized after the user stops speaking. In this paper, we develop a first-pass Recurrent Neural Network Transducer (RNN-T) model and a second-pass Listen, Attend, Spell (LAS) rescorer that surpasses a conventional model in both quality and latency. On the quality side, we incorporate a large number of utterances across varied domains to increase acoustic diversity and the vocabulary seen by the model. We also train with accented English speech to make the model more robust to different pronunciations. In addition, given the increased amount of training data, we explore a varied learning rate schedule. On the latency front, we explore using the end-of-sentence decision emitted by the RNN-T model to close the microphone, and also introduce various optimizations to improve the speed of LAS rescoring. Overall, we find that RNN-T+LAS offers a better WER and latency tradeoff compared to a conventional model. For example, for the same latency, RNN-T+LAS obtains a 8% relative improvement in WER, while being more than 400-times smaller in model size.

Liang Qiao, Sanli Tang, Zhanzhan Cheng et al.

Many approaches have recently been proposed to detect irregular scene text and achieved promising results. However, their localization results may not well satisfy the following text recognition part mainly because of two reasons: 1) recognizing arbitrary shaped text is still a challenging task, and 2) prevalent non-trainable pipeline strategies between text detection and text recognition will lead to suboptimal performances. To handle this incompatibility problem, in this paper we propose an end-to-end trainable text spotting approach named Text Perceptron. Concretely, Text Perceptron first employs an efficient segmentation-based text detector that learns the latent text reading order and boundary information. Then a novel Shape Transform Module (abbr. STM) is designed to transform the detected feature regions into regular morphologies without extra parameters. It unites text detection and the following recognition part into a whole framework, and helps the whole network achieve global optimization. Experiments show that our method achieves competitive performance on two standard text benchmarks, i.e., ICDAR 2013 and ICDAR 2015, and also obviously outperforms existing methods on irregular text benchmarks SCUT-CTW1500 and Total-Text.

Yuan Shangguan, Jian Li, Qiao Liang et al.

While most deployed speech recognition systems today still run on servers, we are in the midst of a transition towards deployments on edge devices. This leap to the edge is powered by the progression from traditional speech recognition pipelines to end-to-end (E2E) neural architectures, and the parallel development of more efficient neural network topologies and optimization techniques. Thus, we are now able to create highly accurate speech recognizers that are both small and fast enough to execute on typical mobile devices. In this paper, we begin with a baseline RNN-Transducer architecture comprised of Long Short-Term Memory (LSTM) layers. We then experiment with a variety of more computationally efficient layer types, as well as apply optimization techniques like neural connection pruning and parameter quantization to construct a small, high quality, on-device speech recognizer that is an order of magnitude smaller than the baseline system without any optimizations.

Tara N. Sainath, Ruoming Pang, David Rybach et al.

The requirements for many applications of state-of-the-art speech recognition systems include not only low word error rate (WER) but also low latency. Specifically, for many use-cases, the system must be able to decode utterances in a streaming fashion and faster than real-time. Recently, a streaming recurrent neural network transducer (RNN-T) end-to-end (E2E) model has shown to be a good candidate for on-device speech recognition, with improved WER and latency metrics compared to conventional on-device models [1]. However, this model still lags behind a large state-of-the-art conventional model in quality [2]. On the other hand, a non-streaming E2E Listen, Attend and Spell (LAS) model has shown comparable quality to large conventional models [3]. This work aims to bring the quality of an E2E streaming model closer to that of a conventional system by incorporating a LAS network as a second-pass component, while still abiding by latency constraints. Our proposed two-pass model achieves a 17%-22% relative reduction in WER compared to RNN-T alone and increases latency by a small fraction over RNN-T.

Yanzhang He, Tara N. Sainath, Rohit Prabhavalkar et al.

End-to-end (E2E) models, which directly predict output character sequences given input speech, are good candidates for on-device speech recognition. E2E models, however, present numerous challenges: In order to be truly useful, such models must decode speech utterances in a streaming fashion, in real time; they must be robust to the long tail of use cases; they must be able to leverage user-specific context (e.g., contact lists); and above all, they must be extremely accurate. In this work, we describe our efforts at building an E2E speech recognizer using a recurrent neural network transducer. In experimental evaluations, we find that the proposed approach can outperform a conventional CTC-based model in terms of both latency and accuracy in a number of evaluation categories.