The objective in Track 1 is to develop a high-quality audio denoising solution that operates efficiently on a neuromorphic system. The solution must follow the following structure: noisy audio is encoded into a form suitable for processing on a neuromorphic system, processed on a simulated neuromorphic system, and decoded into a clean output audio waveform. Solutions are evaluated by audio quality metrics and proxy computational resource usage metrics and are subject to a minimum audio quality and maximum latency (real-time) requirement. The neuromorphic denoiser is not required to run on actual neuromorphic hardware, but rather will be simulated on conventional hardware (CPU). The encoder and decoder run on CPU.
Track 1 provides freedom to explore a wide range of neuromorphic denoising solutions, without the need to demonstrate the solutions on actual neuromorphic hardware; this track is intended for rapid development and potentially to inspire future neuromorphic hardware features.
Coming soon.