Audio & Speech

SELEBI algorithm fixes 'percussion smearing' in audio time-stretching

Researchers' new method uses adaptive windowing to preserve drum hits when slowing down or speeding up audio.

Deep Dive

Researchers Natsuki Akaishi, Nicki Holighaus, and Kohei Yatabe developed SELEBI, a new phase vocoder algorithm for audio time-stretching. It uses a Nonstationary Gabor Transform to dynamically adapt analysis window lengths, assigning short windows to percussive sounds. This directly addresses the 'percussion smearing' artifact that degrades drum quality in conventional methods. The approach ensures temporal consistency between magnitude and phase, maintains perfect reconstruction, and produces more natural-sounding results in experiments.

Why It Matters

Delivers cleaner, artifact-free time-stretching for music producers, podcast editors, and audio engineers working with rhythmic content.

📬 Get the top 10 AI stories daily