Search

To solve the problems that the Attention's overly flexible alignment was poorly adaptable in complex environments and the language features were not fully utilized by simple end-to-end models, a speech recognition method was investigated based on multi-task loss with additional language model. By analyzing the characteristics of the speech signal, the features containing more information were selected in the training. Based on the Attention-based Conformer end-to-end model, the model was trained using multi-task loss of CTC loss assisted pure Conformer (Attention), and the Conformer-CTC speech recognition model was obtained. Based on the Conformer-CTC model, by analyzing and comparing the characteristics and effects of some language models, the Transformer language model was added to the training of the above model through re-scoring mechanism, and the Conformer-CTC-Transformer speech recognition model was obtained. The experiments on the above model were completed on the AISHELL-1 data set. The results show that compared with the pure Conformer (Attention) model, the character error rate (CER) of the Conformer-CTC model on the test set is reduced by 0. 49%, and the CER of the Conformer-CTC-Transformer model on the test set is reduced by 0.79% compared with the Conformer- CTC model. The adaptability of Attention alignment in complex environments can be improved by CTC loss, and after re-scoring the Transformer-CTC model with the Transformer language model, the recognition accuracy can be increased by 0. 30% again. Compared with some existing end-to-end models, the recognition effect of the Conformer-CTC-Transformer model is better, indicating that the model has certain effectiveness. [ABSTRACT FROM AUTHOR]

Recently, deep neural network model has become a hot research object in the field of speech recognition. However, the deep neural network relies on huge parameters and computational overhead, the excessively large model size also increases the difficulty of its deployment on edge devices. Aiming at the above problems, this paper proposed a lightweight speech recognition model based on Transformer. This method used depthwise separable convolution to obtain the feature information. Secondly, this method constructed a two half-step feed-forward layers, namely Macaron-N et, and introduced the lowrank matrix factorization to realize the model compression. Finally, it used a sparse attention mechanism to improve the training speed and decoding speed of the model. It tested on the Aishell-1 and aidatang_200zh datasets. The experimental results show that compared with Open-Transformer, the word error rate and real time factor of LM-Transformer decrease by 19. 8% and 32.1%, respectively. [ABSTRACT FROM AUTHOR]

To improve the performance of multi-accent Mandarin speech recognition task,this paper presented a method for hybrid end-to-end automatic speech recognition(ASR) by combining CTC and multi-head attention by using a multiobjective training and joint decoding.The analysis shows that hybrid model with lower CTC weight and deeper encoder layers performance better learning capacity.And it trained a very deep models with up to 48 layers for encode-decoder architecture,which outperforms all previous end-to-end ASR approaches on Aidatatang 200 h multi-accent dataset,achieves 5.6% character error rate(CER) and 26.2% sentence error rate(SER).The experiment proves that the recognition rate of the end-to-end model proposed exceeds the general end-to-end model,and it has certain advancedness in solving the Mandarin recognition with accents. [ABSTRACT FROM AUTHOR]

This paper proposed a framework which attempted to incorporate articulatory information into the stochastic segment model based on Mandarin speech recognition system. According to the characteristics of the stochastic segment model, it used hierarchical artificial neural network to obtain the posterior probability of speech signal belonging to the phonemes. Then, it integrated the posterior probability into the stochastic segment model system in the first search process. Experiments conducted on "863-test" set show that about 5.93% relative improvement could be achieved in the recognition accuracy. Thus, it demonstrates the feasibility of the method. [ABSTRACT FROM AUTHOR]