This website requires JavaScript.

The Intel Neuromorphic DNS Challenge

Jonathan TimchekSumit Bam ShresthaDaniel Ben Dayan Rubin ...+4 Mike Davies
Mar 2023
A critical enabler for progress in neuromorphic computing research is theability to transparently evaluate different neuromorphic solutions on importanttasks and to compare them to state-of-the-art conventional solutions. The IntelNeuromorphic Deep Noise Suppression Challenge (Intel N-DNS Challenge), inspiredby the Microsoft DNS Challenge, tackles a ubiquitous and commercially relevanttask: real-time audio denoising. Audio denoising is likely to reap the benefitsof neuromorphic computing due to its low-bandwidth, temporal nature and itsrelevance for low-power devices. The Intel N-DNS Challenge consists of twotracks: a simulation-based algorithmic track to encourage algorithmicinnovation, and a neuromorphic hardware (Loihi 2) track to rigorously evaluatesolutions. For both tracks, we specify an evaluation methodology based onenergy, latency, and resource consumption in addition to output audio quality.We make the Intel N-DNS Challenge dataset scripts and evaluation code freelyaccessible, encourage community participation with monetary prizes, and releasea neuromorphic baseline solution which shows promising audio quality, highpower efficiency, and low resource consumption when compared to MicrosoftNsNet2 and a proprietary Intel denoising model used in production. We hope theIntel N-DNS Challenge will hasten innovation in neuromorphic algorithmsresearch, especially in the area of training tools and methods for real-timesignal processing. We expect the winners of the challenge will demonstrate thatfor problems like audio denoising, significant gains in power and resources canbe realized on neuromorphic devices available today compared to conventionalstate-of-the-art solutions.