Velvet noise is a sparse pseudo-random signal characterized by alternating positive and negative impulses (or sometimes a single polarity) positioned irregularly in time.
Understanding Velvet Noise
Unlike white noise, which has a uniform distribution of energy across all frequencies, velvet noise is deliberately sparse. This sparsity is key to its unique properties. Its creation involves randomly placing a predetermined number of impulses within a given time frame. The impulses are often of alternating polarity (+1 and -1), though sometimes the construction only uses one polarity (all +1, for example).
Key Characteristics:
- Sparsity: The signal contains relatively few non-zero values (impulses) compared to the overall length of the signal.
- Pseudo-Randomness: The positions of the impulses are determined by a pseudo-random number generator, meaning they appear random but are reproducible if the same seed is used.
- Alternating Polarity (Usually): Impulses typically alternate between positive and negative values, although this isn't always a requirement.
Applications of Velvet Noise:
Velvet noise finds applications in various areas of audio processing and beyond:
-
Late Reverberation Modeling: Due to its sparse and uncorrelated nature, velvet noise is used in creating realistic-sounding late reverberation effects in audio processing. It can efficiently simulate the complex reflections and diffusion that occur in acoustic spaces.
-
Decorrelation: Velvet noise can be used to decorrelate audio signals, reducing unwanted correlations between channels and enhancing the spatial perception of audio. This is useful in stereo widening and surround sound techniques.
-
Speech Generation: In some speech synthesis methods, velvet noise can be used as an excitation signal to generate speech-like sounds. The sparse nature of the noise allows for more efficient processing.
-
Extending Signals: Because velvet noise is not perceived like static noise it can sometimes be used to artificially extend the audible duration of decaying sounds, like a cymbal crash.
Example:
Imagine a 1-second audio signal. Instead of filling every sample with a random value (like white noise), velvet noise might contain only 100 impulses, randomly placed throughout that second, with alternating positive and negative values.
Benefits:
-
Computational Efficiency: The sparsity of velvet noise leads to reduced computational cost compared to dense signals like white noise. This is especially advantageous in real-time audio processing applications.
-
Perceptual Transparency: Well-designed velvet noise can be perceptually transparent, meaning it is not easily audible as noise itself, but rather contributes to the overall effect (e.g., in reverberation).
