Related Experiment Videos
Squared error as a measure of perceived phase distortion.
Harald Pobloth1, W Bastiaan Kleijn
1Department of Signals, Sensors, and Systems KTH (Royal Institute of Technology), 100 44 Stockholm, Sweden. harald.pobloth@s3.kth.se
The Journal of the Acoustical Society of America
|August 29, 2003
Summary
The squared error measure for speech phase distortion is perceptually relevant for small errors but not large ones. Using this measure can limit perceived distortion improvements in speech coding.
Area of Science:
- Speech processing
- Auditory modeling
- Signal processing
Background:
- Fourier phase spectrum distortions can impact speech perception.
- Auditory models provide frameworks for understanding perceptual relevance.
Purpose of the Study:
- To evaluate the perceptual relevance of squared error for speech Fourier phase distortions.
- To analyze phase vector quantizer performance and error-distortion relationships.
Main Methods:
- Utilized two established auditory models.
- Investigated squared error against perceptual distortion measures.
- Conducted empirical rate-perceptual distortion analyses and listening tests.
Main Results:
- Squared error correlates well with perceptual measures for small distortion values.
- For larger errors, squared error increase does not consistently raise perceptual distortion.
- Average perceived distortion plateaus with increasing codebook size when using squared error.
Conclusions:
- Squared error is a limited perceptual measure for speech phase distortion, especially at higher error levels.
- Current speech coding approaches using squared error may not optimize perceptual quality effectively.