Toneboosters releases TB Dither plug-in for Mac and Windows 17/02/14
Toneboosters tells us that TB Dither is the world's first quantization and noise shaping plugin that allows the design of your own noise shaping curve – as easy as working with an EQ. Here's the story in their own words:
TB Dither is a plugin designed to modify the bit depth of audio signals, for example when authoring a CD or for archival purposes, with minimum quality degradation. Such process typically involves dithering, quantization, and noise shaping. TB Dither supports industry-standard dithering noise types such as RPDF (rectangular probability density function, 1 LSB wide) and TPDF (triangular probability density function, 2 LSBs wide). A GPDF (Gaussian probability density function) is provided as well.
TB Dither's uniqueness lies in the flexibility to shape and minimize the audibility of noise inherently introduced by bit depth reduction. Instead of providing a very limited set of a few, fixed noise shaping curves, TB Dither allows you to design the spectrum of the quantization noise using familiar tools such as low-shelf, high-shelf and peaking filters, just as any equalizer! This provides an unprecented ability to adjust quantization noise spectra according to the audio content, and envisioned reproduction system(s). If you can work with an EQ, you can work with TB Dither!
To get started, no less than 7 different noise shaping curves are provided and can be recalled from a menu, ranging from threshold-in-quiet curves, inverse dB(A) weighting, inverse ITU-R 468 curves, and several more.
Audibility of sample rate and bit depth reduction According to double-blind tests, the only audible effect when converting high-resolution audio to a sample rate of 44.1 kHz and 16 bits is the injected (dithering and quantization) noise. TB Dither resolves this by decreasing the quantization noise level in the frequency range the human ear is most sensitive to, and thereby increasing the dynamic range for those frequencies beyond the 16-bit limit. The audio example below clearly shows this benefit.
Dithering and information theory When working with TB Dither, there are a couple of things one should know about quantization and the resulting quantization noise. Bit depth reduction will always introduce errors. If approprate dithering is applied, these errors will manifest themselves as spectrally flat (or white) noise. With additional noise shaping techniques, we can change the spectrum of that quantization noise to make it less audible. A very common approach is to reduce the noise level in the frequency region the human hearing system is most sensitive to (1 kHz to 10 kHz, approximately). Such noise shaping is, however, subject to information-theoretic limitations, the most important one being that the total amount of noise cannot be reduced, only be increased. In other words, decreasing the noise level between 1 and 10 kHz will always result in an increase of noise below 1 kHz and/or above 10 kHz!
Fortunately, as end-user of TB Dither, you do not have to worry about information-theoretic principles. The way it works is as follows. Use the four noise shaping sections to construct a desired noise spectrum, in the same way you operate an equalizer. TB Dither will, in real time, automatically find the best-possible realization of that curve by shifting the curve up and down such that the overall quantization noise level is the lowest possible noise level for that specified curve. An example is given in this screenshot. The colored areas indicate the targeted spectrum, while the white curve shows the resulting noise spectrum, compared to applying no noise shaping at all. In this example, the quantization noise has been reduced by up to 20 dB for frequencies below approximately 10 kHz, at the expense of an increase of noise in the frequency region humans are not so sensitive to.
Support of all sampling rates from 44.1 to 192 kHz
Supports industry-standard RPDF and TPDF dithering noise
Unrivalled flexibility to shape dithering and quantization noise
Suports any bith depth between 8 and 24 bits
Dedicated modes to listen to quantization noise only