IP3 2024 – Speech Noise Suppression (Lot 15389)

This patent is generally related to an audio signal processing system to improve the audio quality of voice calls in a communication device. Disclosed is a technique to improve the quality of a voice call consisting of a single microphone for receiving a near-end voice signal, a near-end noise signal, and an input audio signal from the microphone. The system reclassifies the input audio signal as a noise-only signal by determining that an active-speech signal is not present and converting the input audio signal into an input audio spectrum. Also disclosed techniques to update a noise spectrum with the input audio spectrum if the signal is classified as a noise-only signal and compare the input audio spectrum with the noise spectrum. The system updates a noise-reduced spectrum by multiplying a predetermined gain shaping factor with the input audio spectrum if the input audio spectrum is within or below a predetermined threshold in the noise spectrum. Further disclosed are techniques to convert the noise-reduced spetrum to a time-domain noise-reduced audio signal and generate a high-quality noise-reduced audio signal for a user. The technology may be implemented in noise suppressor systems, background noise reduction systems, speech recognition systems, etc.

How Can We Help?