A malleable audio representation for data compression
Scott Levine · The Journal of the Acoustical Society of America · 2000
The boundaries between audio data compression and audio synthesis/modifications are blurring as research in both fields progresses. New results using a sines + transients + noise parametric representation allows for coding gains as efficient as the most recent nonparametric transform coding schemes. Because the audio input is decomposed into separate sinusoidal, transient, and noiselike signals, aggressive quantization can be performed independently on each signal in a perceptually meaningful manner. While the sines + transients + noise representation allows for good compression rates, it also allows for high-quality time-scale modification in the compressed domain. The multiresolution sinusoidal modeling and the Bark-band noise modeling utilized in this representation have been used in both the music synthesis and audio/speech coding arenas. The transient modeling in this representation uses similar algorithms to those used in transform coding systems.