Comparable Study Of Modeling Units For End-To-End Mandarin Speech Recognition

By Wei Zou, Dongwei Jiang, Shuaijiang Zhao, Xiangang Li. 2018

35 citations recorded by the source corpus.

End-to-end speech recognition has become increasingly popular in Mandarin speech recognition and achieved delightful performance. Mandarin is a tonal language which is different from English and requires special treatment for the acoustic modeling units. There have been several different kinds of modeling units for Mandarin such as phoneme, syllable and Chinese character. In this work, we explore two major end-to-end models: connectionist temporal classification (CTC) model and attention based encoder-decoder model for Mandarin speech recognition. We compare the performance of three different scaled modeling units: context dependent phoneme (CDP), syllable with tone and Chinese character. We find that all types of modeling units can achieve similar character error rate (CER) in CTC model and the performance of Chinese character attention model is better than syllable attention model. Furthermore, we find that Chinese character is a reasonable unit for Mandarin speech recognition. On HKUST task, Chinese character attention model achieves a CER of 35.2% and CTC model gets a CER of 35.7%, on internal Callcenter task, CERs are 5.68% and 7.29%, respectively, on internal Reading task, CERs are 4.89% and 5.79%, respectively. Moreover, attention model achieves a better performance than CTC model on these datasets.

Subfields: Artificial Intelligence, Natural Language Processing, Speech & Audio, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Bo Xu
2. Linhao Dong
3. Shiyu Zhou
4. Ne Luo
5. Wubo Li
6. Shuang Xu
7. Wei Zou
8. Dongwei Jiang

In Natural Language Processing

1. Bo Xu
2. Linhao Dong
3. Shiyu Zhou
4. Shuang Xu
5. Yuxuan Hu
6. Xiangang Li
7. Dongwei Jiang
8. Wei Zou

In Speech & Audio

1. Bo Xu
2. Linhao Dong
3. Ne Luo
4. Wubo Li
5. Xiangang Li
6. Wei Zou
7. Dongwei Jiang
8. Shiyu Zhou

In Signal Processing

1. Bo Xu
2. Linhao Dong
3. Shiyu Zhou
4. Ne Luo
5. Wubo Li
6. Shuang Xu
7. Wei Zou
8. Dongwei Jiang

Citations in the corpus

• Improving Transformer-based Speech Recognition Using Unsupervised Pre-training (2019, methodology): “Mandarin characters are used as modeling units in all our experiments [14, 29].”
• A Comparison of Modeling Units in Sequence-to-Sequence Speech Recognition with the Transformer on Mandarin Chinese (2018, background): “What’s more, [13] compared the modeling units of characters and syllables by sequence-to-sequence attention-based models.”
• A Further Study of Unsupervised Pretraining for Transformer Based Speech Recognition (2020, methodology): “Mandarin characters are used as modeling units for HKUST and AISHELL like described in [36], while 2000 BPE subwords [37] are used for experiments on Switchboard.”
• Didispeech: A Large Scale Mandarin Speech Corpus (2020, background): “The DiDiSpeech-2 is established for the purpose of building multi-speaker speech synthesis or ASR system.”