AONDevices unveils low-latency, customisable edge AI speech enhancement

1 min read

AONDevices, a provider of low-power, high-performance on-device AI processors and full stack solutions, has announced the launch of AONDenoise.

This compact, efficient edge AI denoising technology has been designed for applications that require minimal power and latency and is said to be one of the smallest denoising algorithms available, making it suitable for AI speech enhancement in hearing aids, wireless earbuds, smartphones, and wearables.

Traditionally, AI speech enhancement is performed using arrays of multiple microphones, beamforming and large Digital Signal Processing (DSP) algorithms. This method, currently implemented in today’s smartphones and other devices, requires huge amounts of memory and power, which is costly.

Through its AI-based denoiser AONDevices is looking to addressing these concerns.

With just 50K parameters, the AONDenoise solution is one of the most compact denoising algorithms currently on the market and is multiple orders of magnitude smaller than traditional DSP noise reduction methods and at least an order of magnitude smaller than current AI-based denoisers.

The AONDenoise solution addresses the need for customisable noise cancellation in ultra-low power, real-time processing devices, such as chips used for hearing aids. The algorithm can target a wide range of noises, such as chatter in a crowded restaurant, wind noise, dogs barking, babies crying, or keyboards clicking.

The AONDenoise solution could help to revolutionise various applications, including:

  • Hearing Aids: Customisable noise cancellation for improved performance in different environments.
  • Wireless Earbuds: Optimal audio quality in transparency mode, conversation mode, and voice calls.
  • Smartphones: Clearer voice quality during calls, especially in outdoor settings.
  • Podcasting: Crystal-clear audio streaming even in noisy surroundings.

Leveraging AI and deep learning neural networks, the AONDenoise solution can distinguish human speech from background noise. To maximise the efficiency of this neural network, AONDevices has developed a dedicated software tool suite that allows both in-house and customer development to customise and train the system to handle specific noise environments or a broader range of noises.

The AONDenoise solution is now available for integration into third-party processors. For customers seeking dedicated hardware solutions, AONDevices is preparing to launch the denoiser hardware IP and the associated next-generation chip – the AON2100 – in the coming months.