Code-switched English Pronunciation Modeling for Swahili Spoken Term Detection

Kleynhans, Neil; Hartman, William J.; Niekerk, Daniel van; Heerden, Charl van; Schwartz, Rich; Tsakalidis, Stavros; Davel, Marelie H.

doi:10.1016/j.procs.2016.04.040

Cited by 8 publications

(6 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Other measures of performance accuracy include measures like character error rate (CER), Word accuracy (WA), Mixed Error Rate (MER), Phrase Error Rate (PER) and Token Error Rate (TER) both which apply word error rates and character error rates [5,11,21,24,35]. The performance of the most ASR engines is measured in terms of WER and Word Recognition Rate (WRR) [7,17,20,21,24,32,34,36].…”

Section: B Software and Toolkits For Bilingual Asr Development And Ev...mentioning

confidence: 99%

Bilingual Automatic Speech Recognition: A Review, Taxonomy and Open Challenges

et al. 2023

View full text Add to dashboard Cite

In this technological era, smart and intelligent systems that are integrated with artificial intelligence (AI) techniques, algorithms, tools, and technologies, have impact on various aspects of our daily life. Communication and interaction between human and machine using speech becomes increasingly important, since it is an obvious substitute for keyboards and screens in the communication process. Therefore, numerous technologies take advantage of speech such as Automatic Speech Recognition (ASR), where human natural speech for many languages is used as the mean to interact with machines. Majority of the related works for ASR concentrate on the development and evaluation of ASR systems that serve a single language (monolingual) only, such as Arabic, English, Chinese, French, and many others. However, research attempts that combine multiple languages (bilingual and multilingual) during the development and evaluation of ASR systems are very limited. This paper aims to provide comprehensive research background and fundamentals of bilingual ASR, and related works that have combined two languages for ASR tasks from 2010 through 2021. It also formulates research taxonomy and discusses open challenges to bilingual ASR research. Based on our literature investigation, it is clear that bilingual ASR using deep learning approach is highly demanded and is able to provide acceptable performance. In addition, many combinations of two languages such as Arabic-English, Arabic-Malay, and others, are not attempted yet by the research community, which can open new research opportunities. Finally, it is clear that ASR research is moving towards not only bilingual ASR, but also multilingual ASR.

show abstract

Section: B Software and Toolkits For Bilingual Asr Development And Ev...mentioning

confidence: 99%

Bilingual Automatic Speech Recognition: A Review, Taxonomy and Open Challenges

et al. 2023

View full text Add to dashboard Cite

show abstract

“…4) The Mel-frequency cepstral coefficient features of every audio frame are input into the LSTM in order for forward calculation [17]:…”

Section: B Speech Recognition-based Spoken Pronunciation Scoring Modulementioning

confidence: 99%

Design of a Speaking Training System for English Speech Education using Speech Recognition Technology

He¹

2022

IJACSA

View full text Add to dashboard Cite

A good English speaking training system can provide an aid to the learning of English. This paper briefly introduced the English speaking training system and described the speaking training scoring and pronunciation resonance peak display modules in the system. The speaking training scoring module scored pronunciation with the Long Short-Term Memory (LSTM). The pronunciation resonance peak display module extracted the resonance peak with Fourier transform and visualized it. Finally, the speaking scoring module, the pronunciation resonance peak display module, and the effect of the whole system in improving students' speaking pronunciation was tested. The results showed that the LSTM-based speaking scoring algorithm had highest scoring accuracy than pattern matching and the recurrent neural network (RNN) algorithm, and its accuracy was 95.21% when scoring the LibriSpeech dataset and 90.12% when scoring the local English dataset. The pronunciation resonance peak display module displayed the change of mouth shape before and after training, and the pronunciation after training was closer to the standard pronunciation. The P value in the comparison of the speaking level before and after training with the system was 0.001, i.e., the difference was significant, which indicated that the students' English speaking proficiency significantly improved.

show abstract

“…In this approach, impact of a bilingual Deep Neural Network (DNN) in the context of code-switching speech, an automatic speech recognition is investigated. The detection of language boundaries for code-switch speech utterances for languages such as Swahili and English have been proposed in [6]. In [7] an approach to identify the language boundary in codeswitching utterances with the use of bi-phone probabilities is explored.…”

Section: Related Workmentioning

confidence: 99%

SVM Based Language Diarization for Code-Switched Bilingual Indian Speech Using Bottleneck Features

Spoorthy¹,

Thenkanidiyoor²,

Dinesh

2018

6th Workshop on Spoken Language Technologies for Under-Resourced Languages (SLTU 2018)

View full text Add to dashboard Cite

This paper proposes an SVM-based language diarizer for codeswitched bilingual Indian speech. Code-switching corresponds to usage of more than one language within a single utterance. Language diarization involves identifying code-switch points in an utterance and segmenting it into homogeneous language segments. This is very important for Indian context because every Indian is at least bilingual and code-switching is inevitable. For building an effective language diarizer, it is helpful to consider phonotactic features. In this work, we propose to consider bottleneck features for language diarization. Bottleneck features correspond to output of a narrow hidden layer of a multilayer neural network trained to perform phone state classification. The studies conducted using the standard datasets have shown the effectiveness of the proposed approach.

show abstract

Code-switched English Pronunciation Modeling for Swahili Spoken Term Detection

Cited by 8 publications

References 2 publications

Bilingual Automatic Speech Recognition: A Review, Taxonomy and Open Challenges

Bilingual Automatic Speech Recognition: A Review, Taxonomy and Open Challenges

Design of a Speaking Training System for English Speech Education using Speech Recognition Technology

SVM Based Language Diarization for Code-Switched Bilingual Indian Speech Using Bottleneck Features

Contact Info

Product

Resources

About