Distortion measurement for noise suppression system
View Patent ↗The present technology measures distortion introduced by a noise suppression system. The distortion may be measured as the difference between a noise-reduced speech signal and an estimated idealized noise reduced reference (EINRR). The EINRR may be determined from a speech component and noise component that are pre-processed, and the EINRR may be used with masks associated with energies lost and added in the speech component and noise component. The EINRR may be calculated on a time varying basis.
1. A method for measuring distortion in a noise-reduced signal, comprising:
receiving a noise component which does not contain speech;
receiving a speech component which does not contain noise;
receiving a combined signal by a noise reduction module, the combined signal formed from the combination of the noise component and the speech component;
generating a noise reduced signal by the noise reduction module, the noise reduction module performing noise reduction to the combined signal to generate the noise reduced signal;
constructing an estimated idealized noise reduced reference from the noise component, the speech component and the noise-reduced signal; and
comparing the noise-reduced signal and the estimated idealized noise reduced reference to calculate a measure of distortion produced by the noise reduction module, the distortion calculated as at least one of the voice energy added, voice energy lost, noise energy added, and noise energy lost in the noise-reduced signal.
2. The method of claim 1 , wherein the estimated idealized noise reduced reference is constructed from a speech gain estimate and noise reduction gain estimate that are time variant.
3. The method of claim 1 , further comprising applying a bandwidth limited gain to the speech signal and the noise signal before constructing an estimated idealized noise reduced reference.
4. The method of claim 1 , further comprising applying a frequency weighted gain to the at least one of the voice energy added, voice energy lost, noise energy added, and noise energy lost.
5. The method of claim 1 , wherein constructing includes applying an estimated speech gain to the speech component.
6. The method of claim 1 , wherein constructing includes applying an estimated noise reduction gain to the noise component.
7. The method of claim 1 , wherein calculating includes:
creating a mask for the at least one of the voice energy added, voice energy lost, noise energy added, and noise energy lost; and
combining the difference of the mask and the estimated idealized noise reduced reference.
8. The method of claim 1 , further comprising mapping the at least one of the voice energy added, voice energy lost, noise energy added, and noise energy lost in the noise-reduced signal to a predicted speech quality mean opinion score.