Reports on Progress:

Predictor A

A large data base of subjective responses gathered in the course of the project was used to model the way listeners assess audio quality. Three different modelling tools were developed:

  1. Predictor A
  2. Predictor B
  3. Quality Adviser

The tools were calibrated using the data base of subjective responses collected in the course of the project.

Predictor A was developed to predict the audio quality as a function of bandwidth. Its working principle can be demonstrated using the following examples.

Example 1

figure: Spat

In this example we simulate that the bandwidth of all channels of a 5-channel audio recording is limited down to 3.5 kHz (telephone bandwidth). Not surprisingly, the predicted quality is Bad. Please note, that in this example, we predict the quality of band-limited recording having an F-B spatial audio characteristic (for example traditional classical music recording with an orchestra in the front and some reverberation in the rear). For more information about typical audio scenes in multichannel audio recordings click here.

Example 2

figure: Spat

In this example we simulate that the bandwidth of a 5-channel audio recording is limited down to 17 kHz. Please note that the predicted audio quality is Excellent and it is approaching 100, which means that the perceptual difference between bandwidth of 20 kHz and the simulated bandwidth of 17 kHz is almost imperceptible.

Example 3

figure: Spat

In this example we try to predict the audio quality of a 5-channel recording in which the front left and right channels have the full bandwidth (20 kHz), the bandwidth of the centre channel is limited to 15 kHz, and the bandwidth of the surround channels are limited to 7 kHz. Surprisingly the predicted audio quality is still Excellent. This can be explained by the fact that in this particular case we predicted the effect of band-limiting of an audio recording having an F-B spatial characteristic (surround channels contain mainly reverberations). According to the results of the subjective tests, the deterioration of quality due to band-limiting of the surround channels reproducing only reverberations is not very significant. On the contrary, the reduction of audio quality caused by the limitation of bandwidth of the surround channels reproducing the foreground content is much more severe (see Example 4 below)

Example 4

figure: Spat

The current version of the Predictor A also allows the user to predict the quality of band-limited audio recordings from the movies including a dialogue in the centre channel (in this case the user is asked to tick the radio button Yes in the Dialogue box).