Getting your Trinity Audio player ready...
|
At the Singapore Institute of Technology (SIT), Professor Ian McLoughlin and his team are pioneering an innovative approach to improving speech clarity using artificial intelligence. Over the past two years, they have developed advanced AI frameworks designed to eliminate background noise from recorded speech, leaving only clear and intelligible voice. This technology, known as deep denoising, represents a significant advancement in audio processing and promises to transform communication in noisy environments.
Clear communication in noisy settings, like a phone call on a crowded train or in a busy café, is often disrupted by background noise, causing frustration. Traditional signal processing methods address this but struggle with unpredictable sounds. AI’s ability to manage these challenges makes it a game-changer for enhancing audio clarity.
Deep denoising uses deep learning models to separate speech from ambient noise during calls or recordings. By training AI to identify and remove various noise types while enhancing spoken words, this technology filters out distractions, delivering clear speech. Professor McLoughlin notes that while traditional methods handle predictable noise, AI’s ability to manage unpredictable sounds significantly improves audio quality.
The development of this technology involved a rigorous process. The research team trained multiple AI models using over 50,000 recordings of noisy speech, paired with clean, noise-free samples. This iterative process, known as back propagation, gradually refined the models’ ability to produce clear speech from noisy input. The AI frameworks were continuously tested and trained until they could reliably transform noisy speech into clear, intelligible audio.
This project, a collaboration between SIT and a Taiwanese electroacoustic company with AI Singapore’s support, involved developing nearly a hundred AI frameworks. The final model was optimised for compact, embedded systems, ensuring real-time performance without compromising quality. This allowed the technology to be used in various applications, from consumer electronics to professional communication tools.
One major challenge was adapting the AI to function efficiently on tiny embedded systems, like those in portable audio devices. While it performed well on desktop computers with powerful GPUs, scaling it down for smaller devices required significant adjustments.
The team simplified and rewrote several mathematical equations within the AI to reduce the number of instructions needed to denoise speech, a process Professor McLoughlin likened to “performing mathematical tricks” to maintain performance while reducing computational complexity.
The result is an AI-powered denoising system capable of operating in real-time with minimal latency, meaning the delay between input and output is imperceptible. This marks a major step forward in speech and audio technology, where the focus has shifted from merely making speech intelligible to improving the overall quality of sound.
Professor McLoughlin, who has worked in speech and audio technology since 1991, has seen the field evolve rapidly, particularly with the rise of AI. In the early days, the primary concern was ensuring that speech was intelligible, especially in critical situations like emergency calls. Today, the emphasis is on enhancing the quality of sound, making it clear and pleasant to listen to.
The potential applications for this technology are vast. Professor McLoughlin is in discussions with industry partners, including rail companies and manufacturers of emergency communication equipment, to explore how SIT’s AI-based denoising technology can be licensed and adopted. If successful, this technology could greatly improve communication in noisy environments, making it a valuable asset across various sectors.
As SIT continues to refine and advance this technology, the future of clear, intelligible speech in noisy environments looks promising. The AI-powered denoising system developed by Professor McLoughlin and his team has the potential to revolutionise global communication by significantly improving sound clarity. This technology could transform how people experience and interact with audio in various aspects of daily life.