A noise-based rounding method used to add a tiny amount of controlled noise to a digital audio file to make other, more objectionable errors less obvious and/or to convert from one word size down to a smaller word, e.g., from 24-bit resolution to 16-bits. Redithering is a dithering process used in digital-to-digital signal processing to distinguish it from the dithering process used during the original A/D conversion. There are several ways to dither:
- Add white noise at about half or one-third the value of the LSB, or half the level that a system can transmit. Thus, for a 16-bit converter encoding a range of 2V, the LSB is equivalent to a difference of 30.5µV (2÷65,536V), yielding dithered noise of about 10-15µV;
- Second-Order dither in which the dither signal (white noise) is processed by a highpass filter to remove low-frequency components. This makes the noise less apparent to the ear, since humans are less sensitive to high-frequency noise than other levels;
- Noise-shaping in which the dither signal is run through a set of filters to provide the most energy in regions where the ear is the least sensitive;
- (Triangular Probability Density Function) where before the noise is shaped, it has a different spectral content than ordinary white noise. It has better noise modulation performance, which is how the noise affects the signal itself;
- Sony’s SBM (Super Bit-Mapping) where the audio is run through a processor with an algorithm that maps a series of higher-resolution samples to a series of lower-resolution samples;
- The Apogee UV-22, which is not really dither. Instead, it uses a periodic signal centered around 22kHz that has good performance in terms of audibility and noise modulation. Placing the signal this high in the audio spectrum makes it very difficult to hear and produces fewer effects on the character of the signal.