Error correction in automatic speech recognition (ASR) aims to correct those incorrect words in sentences generated by ASR models. Since recent ASR models usually have low word error rate (WER), to avoid affecting originally correct tokens, error correction models should only modify incorrect words, and therefore detecting incorrect words is important for error correction. Previous works on error correction either implicitly detect error words through target-source attention or CTC (connectionist temporal classification) loss, or explicitly locate specific deletion/substitution/insertion errors. However, implicit error detection does not provide clear signal about which tokens are incorrect and explicit error detection suffers from low detection accuracy. In this paper, we propose SoftCorrect with a soft error detection mechanism to avoid the limitations of both explicit and implicit error detection. Specifically, we first detect whether a token is correct or not through a probability produced by a dedicatedly designed language model, and then design a constrained CTC loss that only duplicates the detected incorrect tokens to let the decoder focus on the correction of error tokens. Compared with implicit error detection with CTC loss, SoftCorrect provides explicit signal about which words are incorrect and thus does not need to duplicate every token but only incorrect tokens; compared with explicit error detection, SoftCorrect does not detect specific deletion/substitution/insertion errors but just leaves it to CTC loss. Experiments on AISHELL-1 and Aidatatang datasets show that SoftCorrect achieves 26.1% and 9.4% CER reduction respectively, outperforming previous works by a large margin, while still enjoying fast speed of parallel generation.

We propose gated language experts and curriculum training to enhance multilingual transformer transducer models without requiring user input for language identification (LID) during inference. Our method incorporates a gating mechanism and LID loss to enable transformer experts to learn language-specific information. Linear experts are applied on joint network to stabilize training. The curriculum training scheme leverages LID to guide gated experts in improving their respective language-specific performance. Experimental results on an English and Spanish bilingual task show significant average relative word error reductions of 12.5 % and 7.3 % compared to the baseline bilingual and monolingual models, respectively. Our models even perform similarly to upper-bound models with oracle LID. Extending our approach to trilingual, quadrilingual, and pentalingual models reveals similar advantages to those seen in the bilingual models, highlighting its ease of extension to multiple languages.

Code-switching speech refers to a means of expression by mixing two or more languages within a single utterance. Automatic Speech Recognition (ASR) with End-to-End (E2E) modeling for such speech can be a challenging task due to the lack of data. In this study, we investigate text generation and injection for improving the performance of an industry commonly-used streaming model, Transformer-Transducer (T-T), in Mandarin-English code-switching speech recognition. We first propose a strategy to generate codeswitching text data and then investigate injecting generated text into T-T model explicitly by Text-To-Speech (TTS) conversion or implicitly by tying speech and text latent spaces. Experimental results on the T-T model trained with a dataset containing 1,800 hours of real Mandarin-English code-switched speech show that our approaches to inject generated code-switching text significantly boost the performance of T-T models, i.e., 16% relative Token-based Error Rate (TER) reduction averaged on three evaluation sets, and the approach of tying speech and text latent spaces is superior to that of TTS conversion on the evaluation set which contains more homogeneous data with the training set.

Scientific workflows are important in modern computational science and are a convenient way to represent complex computations, which are often geographically distributed among several computers. In many scientific domains, scientists use sensors (e.g., edge devices) to gather data such as CO2 level or temperature, that are usually sent to a central processing facility (e.g., a cloud). However, these edge devices are often not powerful enough to perform basic computations or machine learning inference computations and thus applications need the power of cloud platforms to generate scientific results. This work explores the execution and deployment of a complex workflow on an edge-to-cloud architecture in a use case of the detection and classification of plankton. In the original application, images were captured by cameras attached to buoys floating in Lake Greifensee (Switzerland). We developed a workflow based on that application. The workflow aims to pre-process images locally on the edge devices (i.e., buoys) then transfer data from each edge device to a cloud platform. Here, we developed a Pegasus workflow that runs using HTCondor and leveraged the Chameleon cloud platform and its recent CHI@Edge feature to mimic such deployment and study its feasibility in terms of performance and deployment.

Text error correction aims to correct the errors in text sequences such as those typed by humans or generated by speech recognition models.Previous error correction methods usually take the source (incorrect) sentence as encoder input and generate the target (correct) sentence through the decoder.Since the error rate of the incorrect sentence is usually low (e.g., 10%), the correction model can only learn to correct on limited error tokens but trivially copy on most tokens (correct tokens), which harms the effective training of error correction.In this paper, we argue that the correct tokens should be better utilized to facilitate effective training and then propose a simple yet effective masking strategy to achieve this goal.Specifically, we randomly mask out a part of the correct tokens in the source sentence and let the model learn to not only correct the original error tokens but also predict the masked tokens based on their context information.Our method enjoys several advantages: 1) it alleviates trivial copy; 2) it leverages effective training signals from correct tokens; 3) it is a plug-and-play module and can be applied to different models and tasks.Experiments on spelling error correction and speech recognition error correction on Mandarin datasets and grammar error correction on English datasets with both autoregressive and non-autoregressive generation models show that our method improves the correction accuracy consistently 1 .

In the field of Automatic Speech Recognition (ASR), Grammatical Error Correction (GEC) can be used to correct errors in recognition results of ASR systems and whereby it further reduces the word error rate (WER). Most conventional GEC approaches make correction merely based on a single recognition hypothesis from the ASR system, i.e., the best path in the decoding lattice. However, with such limited information, it's common that GEC fails to detect errors in a sentence or even makes wrong corrections. In this paper we propose two novel GEC approaches utilizing N-best hypotheses instead, that's, N-best plain approach and N-best alignment approach. In this way, our novel approaches can access more information than conventional approaches with 1-best-only hypothesis, and thus more accurate corrections can be made. Experiments on an industrial-scale internal dataset and the open-source dataset AISHELL-1 demonstrate the effectiveness of N-best approaches: 1) both N-best plain and N-best alignment approaches achieve remarkable WER reduction. N-best alignment approach achieves the best performance on both datasets, 12.54% WERR on an internal dataset and 27.71% WERR on AISHELL-1. 2) They also outperform the baseline 1-best GEC approach by a large margin.

Error correction techniques have been used to refine the output sentences from automatic speech recognition (ASR) models and achieve a lower word error rate (WER) than original ASR outputs. Previous works usually use a sequence-to-sequence model to correct an ASR output sentence autoregressively, which causes large latency and cannot be deployed in online ASR services. A straightforward solution to reduce latency, inspired by non-autoregressive (NAR) neural machine translation, is to use an NAR sequence generation model for ASR error correction, which, however, comes at the cost of significantly increased ASR error rate. In this paper, observing distinctive error patterns and correction operations (i.e., insertion, deletion, and substitution) in ASR, we propose FastCorrect, a novel NAR error correction model based on edit alignment. In training, FastCorrect aligns each source token from an ASR output sentence to the target tokens from the corresponding ground-truth sentence based on the edit distance between the source and target sentences, and extracts the number of target tokens corresponding to each source token during edition/correction, which is then used to train a length predictor and to adjust the source tokens to match the length of the target sentence for parallel generation. In inference, the token number predicted by the length predictor is used to adjust the source tokens for target sequence generation. Experiments on the public AISHELL-1 dataset and an internal industrial-scale ASR dataset show the effectiveness of FastCorrect for ASR error correction: 1) it speeds up the inference by 6-9 times and maintains the accuracy (8-14% WER reduction) compared with the autoregressive correction model; and 2) it outperforms the popular NAR models adopted in neural machine translation and text edition by a large margin.

We propose a Laplacian structured sparsity model to study computational branding analytics.To do this, we collected customer reviews from Starbucks, Dunkin' Donuts, and other coffee shops across 38 major cities in the Midwest and Northeastern regions of USA.We study the brand related language use through these reviews, with focuses on the brand satisfaction and gender factors.In particular, we perform three tasks: automatic brand identification from raw text, joint brand-satisfaction prediction, and joint brandgender-satisfaction prediction.This work extends previous studies in text classification by incorporating the dependency and interaction among local features in the form of structured sparsity in a log-linear model.Our quantitative evaluation shows that our approach which combines the advantages of graphical modeling and sparsity modeling techniques significantly outperforms various standard and stateof-the-art text classification algorithms.In addition, qualitative analysis of our model reveals important features of the language uses associated with the specific brands.

The ability to accurately track hand movement provides new opportunities for human computer interaction (HCI). Many of today's commercial hand tracking devices based on gloves can be cumbersome and expensive. An approach that avoids these problems is to use computer vision to capture hand motion. We present a complete real-time hand tracking and 3-D modeling system based on a single camera. In our system, we extract feature points from a video stream of a hand to control a virtual hand model with 2-D global motion and 3-D local motion. The on screen model gives the user instant feedback on the estimated position of the hand. This visual feedback allows a user to compensate for the errors in tracking. The system is used for three example applications. The first application uses hand tracking and gestures to take on the role of the mouse. The second interacts with a 3D virtual environment using the 3D hand model. The last application is a virtual DJ system that is controlled by hand motion tracking and gestures.