Speech and Language Technology for Disabled Persons

Stockholm, Sweden
May 31 - June 2, 1993

New Coding Strategy for the Nucleus Speech Processor

Barry Nevison

Cochlear (UK) Ltd, Dockhead, London, UK

Speech coding strategies employed by Nucleus devices were the first to make deliberate and effective use of the formant structure of speech culminating in the current Mini Speech Processor (MSP) using the MULTIPEAK coding strategy. In 1989, preliminary investigations were carried out at the University of Melbourne using a new, portable research device called the Spectral Maxima Sound Processor (SMSP). The SMSP looks at the running spectral information of the sound signal by using 16 filter banks and extracting the 6 most dominant amplitudes for subsequent stimulation. Interim results on 4 subjects indicated that the performance of the SMSP exceeded that of the MSP giving improved recognition of vowels, consonants, monosyllabic words, and key words in sentences. Based upon these designs Cochlear has developed its own "spectral maxima" coding strategy called SPEAK (Spectral Peak). Results in a pilot study with 5 subjects have showed similar improvements to the SMSP and currently a more extensive trial is under way.

Full Paper

Bibliographic reference.  Nevison, Barry (1993): "New coding strategy for the nucleus speech processor", In SLTDP-93, 23-26.