speech-enhancement

Speech enhancement is a specific signal-processing discipline for improving speech quality. Specific→general.

Speech enhancement algorithms are built specifically to improve speech intelligibility in degraded conditions. The servant technique points to the master goal of making speech understandable; removal of the intelligibility goal removes the purpose for which the enhancement is designed.

Speech-enhancement is a specific kind of speech-processing task. Nearest kind is speech-processing.

Phase vocoder is a time-frequency analysis/synthesis technique built for the sake of speech and audio manipulation — pitch shifting, time stretching, and phase-coherent reconstruction that improve perceived quality and intelligibility. Its designed purpose is to serve speech enhancement workflows that require phase-aware processing. Servant points at master per Law 8d.

Spectral subtraction is a noise reduction technique designed for the purpose of improving speech intelligibility and quality, i.e., serving speech enhancement.

Mask-based enhancement is a specific class of speech enhancement algorithms that use time-frequency masks. Specific→general INSTANCE_OF per Law 9.

Phase retrieval reconstructs phase from magnitude spectrograms to restore perceptual quality of speech signals. It is built for the purpose of serving speech enhancement, servant points at master per Law 8d.

Spectral flatness measure quantifies tonality vs noisiness per frame and is built for the sake of guiding noise suppression and quality assessment in speech enhancement pipelines; it is used as a feature to decide enhancement strength and to evaluate enhancement outcomes. Servant points at master per Law 8d.

Phase coherence is a measurable property developed to assess and improve the perceptual quality and intelligibility of speech signals, i.e., built for the purpose of serving speech enhancement. Servant points at master per Law 8d.

Cepstral subtraction is a noise-suppression technique built specifically for the sake of improving speech intelligibility and quality; it estimates and removes the noise cepstrum to enhance speech. Servant points at master per Law 8d.

Spectral whitening is implemented as a pre-processing step in speech enhancement pipelines to remove spectral coloration and flatten the magnitude spectrum, improving the conditioning of subsequent enhancement estimators. It is built and maintained for the sake of improving speech intelligibility and quality.

Harmonic-to-noise ratio is a voice-quality metric built for the purpose of assessing and controlling periodicity in speech enhancement; it is used to guide dereverberation and noise suppression decisions and to evaluate enhancement quality. Servant points at master per Law 8d.

Griffin-Lim iterative phase reconstruction is built for the sake of recovering perceptually plausible audio waveforms from magnitude spectrograms, a core step in speech enhancement pipelines that need phase-aware reconstruction after mask-based magnitude enhancement. Servant points at master per Law 8d.

Mask-based enhancement applies spectral masks to noisy speech to improve quality and serves speech enhancement directly. The masks are designed to isolate clean speech components from noise, making this technique a core method in the speech enhancement toolkit.

Wiener filter is an optimal linear estimator built for the sake of noise reduction and signal estimation in speech processing pipelines. Its designed purpose is to further speech enhancement operation by minimizing mean-squared error. Servant points at master per Law 8d.