Skyworks’  AI Noise Suppression software claims to deliver noise reduction which  increases speech intelligibility, preserving quality and reducing listener fatigue when using handsets, headsets, conference systems, automotive telematics, wireless microphones and audio-visual systems.

The software scales from desktop CPU/GPU down to edge platforms. It  is trained on a  variety of speech and noise signals and can be implemented on a range of DSP cores and audio frameworks.

IMG_2858.jpeg

The AI is based on a feed-forward neural network that is scalable from enterprise to edge, requiring less than 250 kilobytes of memory and adding no latency when integrated into an existing audio processing path.

The software claims to meet the needs of the market by:
• Achieving subjective and objective test scores in embedded implementations that compete with desktop CPU/GPU solutions while requiring less than 10% CPU load on a HiFi4 core running at 600 MHz.
• The ability to run at 48 kHz, enabling advanced noise reduction on full bandwidth signals in audio/video production and playback.
• Not requiring connection to large cloud servers or tethering to desktop PCs or mobile phones.
• Not interfering with the original speech signal as confirmed by subjective testing based on ITU-T P.808.

The product allows for advanced noise reduction on full bandwidth signals,  opening additional applications in the areas of audio/video production and playback.