Researchers at the University of British Columbia’s Sonic Production, Intelligence, Research, and Applications Lab (SPIRAL) are using advanced machine learning to make networked music performances sound smooth even when data packets are lost in transmission, tells Tech Xplorer. In real-time audio streaming, especially for high-quality music over wireless networks, missing packets produce audible gaps or glitches that break the listening experience. Traditional networking protocols such as UDP send audio data quickly but without error correction, so lost packets can’t be retransmitted and listeners hear silence or distortion.
To address this, SPIRAL researchers are applying neural networks to packet loss concealment. This AI-driven method analyzes the history of the audio stream and generates synthetic audio that closely matches what was lost. Instead of just guessing short silence fillers or simple interpolations, the model creates realistic replacements that blend with ongoing sound, making glitches nearly imperceptible. The approach goes beyond stationary audio signals, common in older concealment techniques, by handling complex, dynamic music and soundscapes where overlapping instruments and varying textures make prediction harder.
The idea grew out of work on Light Up Kelowna, a public art project in which wireless multi-node audio installations revealed packet loss problems in urban radio environments. Undergraduate student Yashvardhan Joshi teamed with Dr. Miles Thorogood to explore how deep learning might not only prevent glitches but offer new creative tools for sound designers. Their research was published in IEEE Access, proposing continuous-time neural network methods tailored for audio concealment.
If extended and refined, this AI approach could help musicians collaborate remotely with fewer artifacts, improve live internet broadcasts, and enhance interactive sound installations. Instead of silence or pops when packets fail, listeners would hear generated audio that feels natural and uninterrupted, a real advance for networked performance and immersive sound experiences.