Topic-Specific Language Model Based on Graph Spectral Approach for Speech Recognition

Takahashi, Satoshi

doi:10.1007/978-0-387-74935-8_35

Cited by 2 publications

(3 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although the methods for constructing a language model for such a small-size corpus, that is a socalled language model adaptation method, have been proposed by many researchers, almost of them are realized by expanding the target corpus with some approaches to increase the statistical reliability. These approaches generally attempt to merge some existing language models to cope with words and phrases not occurred (Out-Of-Vocabulary: OOV) [4] , [5] In recently a number of methods utilizing documents in World Wide Web as language resources for learning have been proposed [6], [7]. Basic strategy is to collect similar documents from Web using words or phrases in speech recognition results and construct a language model with the documents of the retrieval results from Web.…”

Section: A N-gram Language Model Based On Web Retrieval Resultsmentioning

confidence: 99%

N-gram Language Model Based on Multi-Word Expressions in Web Documents for Speech Recognition and Closed-Captioning

Takahashi

Morimoto

2012

2012 International Conference on Asian Language Processing

Self Cite

View full text Add to dashboard Cite

Automatic speech recognition technique is generally used to align the closed caption text to video data. It is important to increase the speech recognition accuracy for the accurate closed-captioning. This paper proposes the method for constructing N-gram language model based on multi word expressions (MWEs) from web retrieval results to improve the speech recognition performance. The web retrieval experiment for examining the distribution of web count numbers for MWEs and the speech recognition experiment for investigating the effectiveness of MWEs are conducted. The experimental results show that the proposed method can improve the recognition performance and the closed-captioning accuracy.

show abstract

Section: A N-gram Language Model Based On Web Retrieval Resultsmentioning

confidence: 99%

N-gram Language Model Based on Multi-Word Expressions in Web Documents for Speech Recognition and Closed-Captioning

Takahashi

Morimoto

2012

2012 International Conference on Asian Language Processing

Self Cite

View full text Add to dashboard Cite

show abstract

“…Man-hung [9] proposed the unsupervised training of a selforganizing unit recognizer based on the HMM, which effectively improved the ability of discovering topic-related keywords through self-organizing unit (SOU) technology. Takahashi [10] proposed a map-based method to specific topic language models. We evaluate the most frequent keyword weights and combine pre-trained speech annotations to clarify the topic definition of the speech.…”

Section: Related Work In Language Model Applied To An Asr Systemmentioning

confidence: 99%

“…In this experiment, Precision, Recall and F1-Score values are taken to evaluate the model comprehensively, as shown in Equations ( 8)- (10).…”

Section: Experimental Environmental and Datamentioning

confidence: 99%

Semi-Supervised Training of Transformer and Causal Dilated Convolution Network with Applications to Speech Topic Classification

Zeng

Zhang

et al. 2021

Applied Sciences

View full text Add to dashboard Cite

Aiming at the audio event recognition problem of speech recognition, a decision fusion method based on the Transformer and Causal Dilated Convolutional Network (TCDCN) framework is proposed. This method can adjust the model sound events for a long time and capture the time correlation, and can effectively deal with the sparsity of audio data. At the same time, our dataset comes from audio clips cropped by YouTube. In order to reliably and stably identify audio topics, we extract different features and different loss function calculation methods to find the best model solution. The experimental results from different test models show that the TCDCN model proposed in this paper achieves better recognition results than the classification using neural networks and other fusion methods.

show abstract

Topic-Specific Language Model Based on Graph Spectral Approach for Speech Recognition

Cited by 2 publications

References 9 publications

N-gram Language Model Based on Multi-Word Expressions in Web Documents for Speech Recognition and Closed-Captioning

N-gram Language Model Based on Multi-Word Expressions in Web Documents for Speech Recognition and Closed-Captioning

Semi-Supervised Training of Transformer and Causal Dilated Convolution Network with Applications to Speech Topic Classification

Contact Info

Product

Resources

About