Our PhD students, HAO Xiang and MA Chenxiang, under the supervision of our Assistant Professor, Dr WU Jibin, have triumphed over their global competitors and won the Track 1 (Algorithmic) of the Intel Neuromorphic Deep Noise Suppression Challenge (Intel N-DNS Challenge). The team will receive a prize valued at US$15,000 (approx. HK$117,000) as recognition of their achievement.
The Intel N-DNS Challenge aimed to push the boundaries of noise reduction technology by encouraging algorithmic innovation that can lead to higher denoising performance while being efficient when implemented as a brain-inspired neuromorphic system. The Intel N-DNS Challenge consists of two tracks. Track 1 aimed to encourage algorithmic innovation and took place between 16 March to 18 September 2023. Track 2 aims to realise the algorithmic innovation in Track 1 on actual neuromorphic hardware and demonstrate a real-time denoising system, which will take place from Oct 2023 to Mar 2024. The winner was selected based on a holistic evaluation that assessed the audio quality, computational resource usage, solution write-up quality, innovativeness, and commercial relevance of the solutions proposed by the competing teams.
HAO and MA’s solution impressed the evaluation committee by demonstrating an audio quality on par with the Intel’s current production-level system, while offering significantly reduced computational cost. The committee also highlighted their achievement of introducing a novel gated spiking neuron model with a learnable and time-varying timescale parameter that experts believe will benefit many different application domains. Additionally, their solution incorporated state-of-the-art techniques from audio denoising, including frequency partitioning, deep filtering, and a loss function to predict DNSMOS using a MetricGAN+ generative adversarial network. Their marvellous work secured them the championship of this challenge.