Related Experiment Videos
Robust speaker's location detection in a vehicle environment using GMM models
Jwu-Sheng Hu1, Chieh-Cheng Cheng, Wei-Han Liu
1Department of Electrical and Control Engineering, National Chiao-Tung University, Hsinchu 300, Taiwan, ROC. jshu@nctu.edu.tw
Summary
This study introduces a new speaker localization method for cars using a single microphone array. It achieves high accuracy in noisy environments, improving speech-based human-computer interaction safety.
Area of Science:
- * Speech processing
- * Acoustics
- * Human-computer interaction
Background:
- * Speech-based human-computer interaction (HCI) is crucial for driver safety.
- * Conventional speaker localization methods lack robustness due to environmental noise and microphone mismatch.
- * Reliable speaker localization is essential for the widespread adoption of speech-based HCI.
Purpose of the Study:
- * To present a novel speaker localization method for vehicle interiors.
- * To demonstrate high accuracy in detecting speaker location using a single linear microphone array.
- * To enhance the reliability of speech-based HCI in driving environments.
Main Methods:
- * Utilized Gaussian mixture models (GMM) to model phase differences between microphones.
- * Modeled complex room acoustics and microphone mismatch effects.
- * Applied the method in near-field, far-field, and noisy conditions within a car cabin.
Main Results:
- * Achieved high accuracy in speaker location detection within a vehicle.
- * Demonstrated robustness in non-line-of-sight scenarios and varying speaker distances.
- * Outperformed the conventional Multiple Signal Classification (MUSIC) method across various signal-to-noise ratios (SNRs).
Conclusions:
- * The proposed GMM-based approach offers a robust solution for speaker localization in cars.
- * This method enhances the reliability of speech-based HCI systems, particularly for driver safety.
- * The technique effectively handles complex acoustic environments and microphone imperfections.