Noise Discrimination
October 27, 2012, AES Convention, San Francisco—Amin Taghipour from the Fraunhofer Institute’s International Audiolabs presented new techniques for psycho-acoustic models that allow better quantization and threshold detection. This perceptual audio coding provides more listener independence.
The threshold for quiet is changed by simultaneous making effects. You need to have dynamic gain and frequency conditions to extract the sound parameter dependencies of frequency and bandwidth. For example, when a tone and a narrow-band burst of noise are mixed, a long duration of the noise exhibits as an obvious amplitude modulation. A short duration burst can be detected at a frequency modulation and is more subtle.
To test this phenomenon , they set up a series of tests to define a sampling window of a pass band and frequency response to measure the perception. Testers were characterized and trained to determine their thresholds. The various window parameters were increased until the subjects could perceive them, and decreased until they could no longer sense the noise.
After the training, the detailed test noise duration was changed up and down to check for upper and lower thresholds. The lower threshold was defined when the subject could not detect any change, and the upper threshold was when they first detected the noise. They have to change the window function but have to phase match any changes to eliminate other cues.
Some general results were that higher frequencies had a lower threshold and the amount of change needed is smaller at lower frequencies. The test frequencies were 375 Hz, 754 Hz, 1456 Hz, and 2658 Hz. The test subjects were expert listeners, recording engineers, etc. The thresholds are correlated with the period of the signal.
The addition of tonal and narrowband noise defines a signal pair. They tried staircase tests to check for frequency dependencies and used the thresholds for some baselines. Their findings indicate a need for frequency adapted temporal resolution in perceptual audio coding.
