ITS staff are developing objective audio quality measurement tools that are based on models of human auditory perception. These tools are intended to provide reliable audio quality measurements in an efficient and timely fashion. They are expected to benefit persons and groups involved in the development, testing, refinement, or standardization of algorithms and equipment that process audio signals by minimizing the number of costly and time consuming formal subjective tests required by those activities. In addition, an objective audio quality measurement tool with sufficiently low complexity and delay could be embedded inside dynamic audio devices in order to optimize user-perceived audio quality under some operating constraints. Examples of these constraints include average bit rate, minimum bit rate, channel error conditions, and packet delay variation.Since speech signals form an important and prevalent class of audio signals, it is expected that objective audio quality measurement tools will be of significant benefit to telecommunications equipment manufacturers, service providers, and end users. In light of the importance of the evolving digital cellular and PCS networks, a special focus involves the development of measurement tools for low-rate digital speech coding algorithms, and digital transmission over time-varying radio channels.
ITS research centers on careful modeling of the human hearing process coupled with attention to the higher-level perception and discrimination processes. Models and test results available in the literature are examined and incorporated when appropriate. The ITS approach is described by the block diagram in the above figure. The perceptual transformation block includes auditory band filtering along with outer and middle ear frequency response characteristics. These operations may be modeled by passing audio signals through the example filter bank shown on the above figure. Additional operations in the perceptual transformation include loudness transformations and frequency and time domain masking effects.
The distance measure block provides a comparison of the two perceptually transformed signals. After some additional processing, an auditory distance value is generated. This value can then be mapped to a user-oriented scale such as the five-point mean opinion score scale. An example of a possible mapping also appears on the above figure.
