Abstract:
In contrast with English, examination of spectrograms of Norwegian vowels indicated that traditional formant measurements [Peterson and Barney, J. Acoust. Soc. Am. 24, 175--184 (1952)] were inadequate to characterize its nine long-vowel categories. Given Norwegian's distinct patterns of formant convergences and attenuations, it is hypothesized that spectral shape is better described by supraformants, second-order acoustic structures shaped by formant structure. Supraformants quantify the continuous spectral shape into distinct peaks associated with each vowel category. Schematic supraformant patterns for Norwegian have been previously reported [J. Acoust. Soc. Am. 98, 2966 (1995)]. Here 387 tokens of Norwegian vowels from 12 native speakers were analyzed (two children, four males, six females). For each vowel token, an LPC plot of peak frequency versus bandwidth was generated with the filter-order set intentionally low in an attempt to extract supraformants instead of formants. The resulting data set for each vowel was subjected to cluster analysis to assist in quantifying the supraformants. Analysis of the vowel tokens for one adult male speaker [Proceedings of the Chicago Linguistic Society 32 (to be published)] demonstrated that observed clusters correspond closely to the posited supraformant patterns. This study extends the analysis method to the entire corpus of data.