Ton G. Wempe

Praat script to detect syllable nuclei and measure speech rate automatically

In this article, we describe a method for automatically detecting syllable nuclei in order to measure speech rate without the need for a transcription. A script written in the software program Praat (Boersma & Weenink, 2007) detects syllables in running speech. Peaks in intensity (dB) that are preceded and followed by dips in intensity are considered to be potential syllable nuclei. The script subsequently discards peaks that are not voiced. Testing the resulting syllable counts of this script on two corpora of spoken Dutch, we obtained high correlations between speech rate calculated from human syllable counts and speech rate calculated from automatically determined syllable counts. We conclude that a syllable count measured in this automatic fashion suffices to reliably assess and compare speech rates between participants and tasks.

Exploring the acoustic vowel space in two-year-old children: results for Dutch and Hungarian

Abstract In the last decade, there has been an increasing interest in exploring patterns of vowel acquisition in young children. Traditionally, researchers attempt to estimate formant values of vowel realizations via acoustic measurements. However, these techniques have yielded questionable results, due primarily to a low sampling rate of the spectrum caused by a high fundamental frequency in young children’s speech. Additionally, the researcher’s knowledge about the intended vowel quality affects the decision pertaining to vowel formants. A frequency domain band filtering analysis method that minimizes the dependence of the results on F 0 is developed to measure the spectral envelopes in children’s utterances automatically, and is applied to existing utterance data sets of Dutch and Hungarian. One further advantage of the current method is that it selects a maximum of 10 measurement points along the length of the utterance. Data reduction of all filter outputs is achieved via Principal Component Analysis (PCA). By using the first 2 eigenvectors, a reference plane is created. The first two eigenvectors account for 54.2 vs. 58.6% in the Dutch and Hungarian data sets, respectively. Next, a common reference plane for Dutch and Hungarian two-year-olds is constructed by balancing the number of utterances that are analyzed per language. Perceptually judged as being correctly pronounced corner vowels of Dutch- and Hungarian-speaking two-year-old boys were mapped onto this common Dutch–Hungarian reference plane. The band filtering method has shown to be robust with regard to signal-to-noise ratios and to the differences in numbers of measurements.

Comparing deaf and hearing Dutch infants: Changes in the vowel space in the first 2 years

The influence of the mother tongue on vowel productions in infancy is different for deaf and hearing babies. Audio material of five hearing and five deaf infants acquiring Dutch was collected monthly from month 5–18, and at 24 months. Fifty unlabelled utterances were digitized for each recording. This study focused on developmental paths in vowel productions. The applied automated band filtering analysis is F0‐independent and results in a spectral envelope, sampled in a 40‐dimensional space. Via a Principal Component Analysis (PCA, data reduction), a vowel space for normally hearing 2‐year‐olds was constructed, enabling the projection of the individual developmental data of the infants in a two‐dimensional reference plane. Comparison of the results for the hearing and the deaf infants over 2 years indicates individual as well as group differences. The amount of hearing loss as well as the communication styles seem to be important factors in explaining differences between the infant categories.