Related Experiment Videos
Speech synthesis by linear interpolation of spectral parameters between dyad boundaries
The Journal of the Acoustical Society of America
|November 1, 1979
Summary
Linear interpolation of spectral parameters for speech synthesis is possible, though it can slightly degrade speech quality. Adding an extra data point within transitions significantly reduces this degradation.
Area of Science:
- Acoustics
- Speech Processing
- Digital Signal Processing
Background:
- Linear prediction analysis allows area parameters to be interpolated with minimal distortion in synthesized speech.
- The efficacy of area parameter interpolation prompts investigation into interpolating other acoustic parameters, like the power spectrum.
Purpose of the Study:
- To investigate the feasibility of linearly interpolating spectral parameters for speech synthesis.
- To assess the impact of spectral parameter interpolation on synthesized speech quality.
Main Methods:
- A speech analysis-synthesis system utilizing spectral parameters (power spectra samples) was simulated.
- Spectral parameters were extracted at dyad boundaries and interpolated for intermediate values.
- Dyad boundaries were manually identified as phoneme transition regions.
Main Results:
- Informal listening tests indicated slight degradation in synthesized speech using linear interpolation of spectral parameters.
- Utilizing an additional interpolation point within dyad boundaries substantially reduced the observed speech degradation.
Conclusions:
- Linear interpolation of spectral parameters is a viable technique for speech synthesis.
- Optimizing interpolation strategies, such as using more data points, can mitigate quality loss.