Chinese dialect identification using segmental and prosodic features

Area of Science:

• Computational Linguistics
• Speech Processing
• Phonetics

Background:

• Accurate Chinese dialect identification is crucial for various applications.
• Existing methods often focus on either segmental or prosodic features independently.

Purpose of the Study:

• To develop and evaluate novel approaches for Chinese dialect identification.
• To investigate the effectiveness of combining segmental and prosodic speech features.
• To improve the accuracy of discriminating between major Chinese dialects in Taiwan.

Main Methods:

• Phonotactic analysis of tokenized broad phonetic sequences for segmental information.
• Development of prosodic models to capture dialect-specific tone sequences.
• A novel approach integrating both broad phonetic and prosodic feature analysis.
• Evaluation using a multispeaker read-speech dataset.

Main Results:

• Segmental analysis alone achieved a certain level of dialect discrimination.
• Prosodic models captured distinctive tone sequence information for dialects.
• The integrated approach combining segmental and prosodic features yielded the highest accuracy.
• The system successfully discriminated among three major Chinese dialects in Taiwan with 93.0% accuracy.

Conclusions:

• The combination of segmental and prosodic features significantly enhances Chinese dialect identification accuracy.
• The proposed integrated approach offers a robust solution for distinguishing between closely related Chinese dialects.
• This research contributes to advancements in automatic speech recognition and language processing for diverse Chinese dialects.