Related Experiment Videos
Discrimination of pathological voices using a time-frequency approach.
Karthikeyan Umapathy1, Sridhar Krishnan, Vijay Parsa
1Department of Electrical and Computer Engineering, The University of Western Ontario, London, ON N6A 5B9, Canada.
IEEE Transactions on Bio-Medical Engineering
|March 12, 2005
Summary
This study introduces a novel joint time-frequency method for analyzing continuous speech to detect pathological voices. This approach achieves 93.4% accuracy, improving upon traditional methods for real-world voice assessment.
Area of Science:
- Speech analysis
- Bioacoustics
- Medical diagnostics
Background:
- Acoustic measures are vital for assessing voice disorders and therapy progress.
- Current methods often rely on sustained vowels, limiting real-world applicability.
- Analyzing continuous speech presents segmentation challenges, hindering research.
Purpose of the Study:
- To develop and validate a novel method for classifying pathological voices using continuous speech signals.
- To overcome the limitations of traditional acoustic analysis by avoiding signal segmentation.
- To enhance the prediction of abnormal voice quality relevant to everyday communication.
Main Methods:
- A joint time-frequency approach was employed for speech signal decomposition.
- An adaptive time-frequency transform algorithm was utilized.
- Features like octave max, octave mean, energy ratio, length ratio, and frequency ratio were extracted and analyzed using statistical pattern classification.
Main Results:
- The proposed method achieved a high classification accuracy of 93.4%.
- Experiments were conducted on a database of 51 normal and 161 pathological talkers.
- The approach successfully classified pathological voices from continuous speech without prior segmentation.
Conclusions:
- The joint time-frequency analysis of continuous speech is effective for pathological voice detection.
- This method offers a more robust and practical approach to voice disorder assessment.
- The findings suggest improved clinical applicability for voice analysis in real-world scenarios.