Two-Stage Training for Chinese Dialect Recognition

Ren, Zongze; Yang, Guofu; Xu, Shugong

doi:10.21437/interspeech.2019-1522

Cited by 16 publications

(7 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Using a Conformer sequence-to-sequence feature extractor, we have successfully trained ASR-based multilingual bottleneck features without explicltly performing forced phone alignment. A similar behavior was observed for a dialect recognition task [26]. We compare these features with a very strong baseline: classical multilingual bottleneck features [11].…”

Section: Discussionsupporting

confidence: 57%

“…With this approach, state-of-the-art language recognition performance has been achieved without defining a frame alignment of phone labels, using only one language for the ASR task [7]. Moreover, for Chinese dialect recognition, the use of the phone forced alignment performed by the acoustic model trained with CTC loss does not improve language recognition performance over the direct use of the sequence-to-sequence bottleneck features [26]. Multi-task training of a joint ASR and LID model has been successfuly performed for English and Hindi corpora [27].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Modeling and Training Strategies for Language Recognition Systems

Duroselle¹,

Sahidullah²,

Jouvet³

et al. 2021

Interspeech 2021

View full text Add to dashboard Cite

Automatic speech recognition is complementary to language recognition. The language recognition systems exploit this complementarity by using frame-level bottleneck features extracted from neural networks trained with a phone recognition task. Recent methods apply frame-level bottleneck features extracted from an end-to-end sequence-to-sequence speech recognition model. In this work, we study an integrated approach of the training of the speech recognition feature extractor and language recognition modules. We show that for both classical phone recognition and end-to-end sequence-to-sequence features, sequential training of the two modules is not the optimal strategy. The feature extractor can be improved by supervision with the language identification loss, either in a fine-tuning step or in a multi-task training framework. Besides, we notice that end-to-end sequence-to-sequence bottleneck features are on par with classical phone recognition bottleneck features without requiring a forced alignment of the signal with target tokens. However, for sequence-to-sequence, the architecture of the model seems to play an important role; the Conformer architectures leads to much better results than the conventional stacked DNNs approach; and can even be trained directly with the LID module in an end-to-end approach.

show abstract

Section: Discussionsupporting

confidence: 57%

Section: Introductionmentioning

confidence: 99%

Modeling and Training Strategies for Language Recognition Systems

Duroselle¹,

Sahidullah²,

Jouvet³

et al. 2021

Interspeech 2021

View full text Add to dashboard Cite

show abstract

“…Adding a large number of pronunciations of each single word also increases the computational cost because adding alternatives increases the search space. Because of all such reasons, recently, the research community has diverted his attention towards DNN-based techniques and, in the recent past, has proposed various DNN-based approaches to achieve accent and dialect identification tasks more effectively [50][51][52][53][54][55].…”

Section: Related Workmentioning

confidence: 99%

A Robust Approach for Speaker Identification Using Dialect Information

Shah

Moinuddin²,

Khan

2022

Applied Computational Intelligence and Soft Computing

View full text Add to dashboard Cite

The present research is an effort to enhance the performance of voice processing systems, in our case the speaker identification system (SIS) by addressing the variability caused by the dialectical variations of a language. We present an effective solution to reduce dialect-related variability from voice processing systems. The proposed method minimizes the system’s complexity by reducing search space during the testing process of speaker identification. The speaker is searched from the set of speakers of the identified dialect instead of all the speakers present in system training. The study is conducted on the Pashto language, and the voice data samples are collected from native Pashto speakers of specific regions of Pakistan and Afghanistan where Pashto is spoken with different dialectal variations. The task of speaker identification is achieved with the help of a novel hierarchical framework that works in two steps. In the first step, the speaker’s dialect is identified. For automated dialect identification, the spectral and prosodic features have been used in conjunction with Gaussian mixture model (GMM). In the second step, the speaker is identified using a multilayer perceptron (MLP)-based speaker identification system, which gets aggregated input from the first step, i.e., dialect identification along with prosodic and spectral features. The robustness of the proposed SIS is compared with traditional state-of-the-art methods in the literature. The results show that the proposed framework is better in terms of average speaker recognition accuracy (84.5% identification accuracy) and consumes 39% less time for the identification of speaker.

show abstract

“…In this work, we did not invest resources to develop an acceptable automatic speech recognition system for each target language with the goal of performing forced phone alignment. We simply trained a unique multilingual end-to-end speech recognition system with the connectionist temporal classification (CTC) loss [21], extending the idea of [22,23]. We used the Conformer architecture [16] with an output layer specific to each target language, with 64 mel-filterbank features as input.…”

Section: Bottleneck Featuresmentioning

confidence: 99%

Language Recognition on Unknown Conditions: The LORIA-Inria-MULTISPEECH System for AP20-OLR Challenge

Duroselle¹,

Sahidullah²,

Jouvet³

et al. 2021

Interspeech 2021

View full text Add to dashboard Cite

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

show abstract

Two-Stage Training for Chinese Dialect Recognition

Cited by 16 publications

References 29 publications

Modeling and Training Strategies for Language Recognition Systems

Modeling and Training Strategies for Language Recognition Systems

A Robust Approach for Speaker Identification Using Dialect Information

Language Recognition on Unknown Conditions: The LORIA-Inria-MULTISPEECH System for AP20-OLR Challenge

Contact Info

Product

Resources

About