Speech GAU: A Single Head Attention for Mandarin Speech Recognition for Air Traffic Control

Abstract

The rise of end-to-end (E2E) speech recognition technology in recent years has overturned the design pattern of cascading multiple subtasks in classical speech recognition and achieved direct mapping of speech input signals to text labels. In this study, a new E2E framework, ResNet–GAU–CTC, is proposed to implement Mandarin speech recognition for air traffic control (ATC). A deep residual network (ResNet) utilizes the translation invariance and local correlation of a convolutional neural network (CNN) to extract the time-frequency domain information of speech signals. A gated attention unit (GAU) utilizes a gated single-head attention mechanism to better capture the long-range dependencies of sequences, thus attaining a larger receptive field and contextual information, as well as a faster training convergence rate. The connectionist temporal classification (CTC) criterion eliminates the need for forced frame-level alignments. To address the problems of scarce data resources and unique pronunciation norms and contexts in the ATC field, transfer learning and data augmentation techniques were applied to enhance the robustness of the network and improve the generalization ability of the model. The character error rate (CER) of our model was 11.1% on the expanded Aishell corpus, and it decreased to 8.0% on the ATC corpus.

Natural Language Processing (NLP) in Aviation Safety: Systematic Review of Research and Outlook into the Future

“…The selected articles show that the end-to-end speech recognition framework is a study trend for better ASR strategies over traditional HMM-based models, especially given the challenges of multilingual recognition, special use of language, and high speech rate. Among the five articles which include ASR for aviation as a primary research objective, three articles adopted an end-to-end framework [3,8,17]. Neural networks-based methods are also widely used in ASR.…”

“…Hence, a structural topic modeling (STM) approach performs better when applied to NTSB reports [2]. NTSB aviation accident reports have non-standard abbreviations that add noise to the model-learning process [3,14]. Similar issues also exist in non-English databases, such as a study on Chinese civil aviation incident reports from the period 2007-2021 by Jiao et al [9] indicates that the experiment was interfered with due to invalid information, such as inconsistent writing norms and standards from different airlines.…”

“…Similar issues also exist in non-English databases, such as a study on Chinese civil aviation incident reports from the period 2007-2021 by Jiao et al [9] indicates that the experiment was interfered with due to invalid information, such as inconsistent writing norms and standards from different airlines. One way to improve the training model performance is to develop a specialized corpus for handling aviation accident reports only [3,14].…”

“…With the rapid development of digital data storing and transmission in the aviation industry, large contextual data such as incident/accident reports [1,2], air traffic communication transcripts [3], passengers' and customers' reviews [4], and manufacturer and maintenance domain-language-based documents [5] have played substantial roles in ensuring safe and efficient operations [6]. However, the analysis of such textual-based data usually requires significant human resource investment [7].…”

Film and television art innovation in network environment by using collaborative filtering recommendation algorithm

“…Finally, a hybrid recommendation model based on ranking learning is proposed, which effectively solves the problem of weight distribution of the results of different recommendation algorithms [2]. Speech enhancement technology also has important applications in cochlear implants and hearing aids [3]. As a preprocessing module of hearing aids, it greatly reduces the noise interference of hearing aids to obtain a higher target speech signal-to-noise ratio [4].…”