| 摘要 |
A method of estimating noise in a speech signal spectrum comprises the steps of determining the noise level at each side of a harmonic peak (12,14,16,42) of the speech signal spectrum; and interpolating the determined noise levels to estimate the noise level during the harmonic peak. The speech signal spectrum will typically comprise a fundamental frequency f0 and a series of regularly spaced harmonic peaks, the noise level may be determined at: points between harmonic peaks; between the first peak and a low frequency endpoint; or between the lat peak and a high frequency endpoint. The width of a harmonic peak may be determined by measuring the signal level of the peak and comparing the level with a threshold, then measuring one frequency bin further away from the peak at a time until the signal level is below a threshold. In an alternative embodiment, the level of noise associated with a harmonic peak is estimated by determining the statistical fit of a harmonic peak to a central lobe, and considering deviations in the fit as the noise level. |
