4th International Conference on Spoken Language Processing

Philadelphia, PA, USA
October 3-6, 1996

A Neural Matrix Model for Active Tracking of Frequency-modulated Tones

Kiyoaki Aikawa, Hideki Kawahara, Minoru Tsuzaki

ATR Human Information Processing Research Laboratories, Kyoto, Japan

Previous paper has shown that the dynamic process of perceiving frequency-modulated (FM) tones can be well simulated by a sound tracking model represented by second-order systems. This paper proposes a new neural operational model for tracking FM tones using a tonotopic neural matrix. The neural cells are connected to build up an auto-regressive architecture. This neural matrix model is characterized by a novel spectral interpolation algorithm based on Lp-norm. The new tracking model enables simultaneous tracking of multiple FM tones. This paper demonstrates that the neural tracking model can successfully simulate perceptual effect such as pitch bounce caused by crossed linear sweep tones. The perceptual overshoot effect is also reasonably explained.

Full Paper

Bibliographic reference.  Aikawa, Kiyoaki / Kawahara, Hideki / Tsuzaki, Minoru (1996): "A neural matrix model for active tracking of frequency-modulated tones", In ICSLP-1996, 578-581.