Anne Bonneau

Perception of the place of articulation of French stop bursts.

This paper deals with the perception of French natural stop bursts and with the role played by the following vowel in this perception. The first experiment verified the ability of listeners to identify long stimuli containing the burst and part of the subsequent vowel. The second and third experiments investigated the identification of stop bursts with and without a priori knowledge of the following vowel. In order to determine the discriminating power of spectral characteristics of the burst, these experiments used fixed-length burst stimuli of 25-ms duration with all traces of vocalic segment cut off. The bursts of initial voiceless stops were extracted from CVC isolated words. The burst provided very reliable information about stop place since listeners identified correctly 87% of the stops, without a priori knowledge of the following vowel. Performance however was context-dependent. Knowing the identity of the vowel led to a slight but statistically significant improvement in stop identification. Nevertheless, the effects of this knowledge were selective and varied with context. Finally, the first experiment proved that a near perfect identification of stops can be achieved only when all main cues (burst spectrum, burst duration, and onset of vocalic formant transitions) were present simultaneously.

LETTER TO THE EDITOR: Identification of vocalic features from French stop bursts

Abstract This paper investigates the perception of vocalic features from French stop bursts. The test corpus was made up of 90 stimuli of 20–25 ms duration extracted from natural monosyllabic words which combined the initial stops /p, t, k/ with the vowels /i, a, u/. In order to cut off all traces of the following vowel, bursts with a duration of less than 20–25 ms were lengthened. Eight listeners identified the vocalic context under a training condition, eight others under a no-training condition. Results showed that the burst onset provided substantial vocalic information (the overall identification rate was 80% under training condition). Under both conditions, the vowel /i/ was clearly identified from /t/ and /k/, and the vowel /u/ was clearly identified from /k/. Knowledge about coarticulation in stop-vowel sequences and its effects on burst acoustics was used in order to find articulatory phenomena and corresponding acoustic patterns which could explain the identification of vowels from bursts.

Identification of vowel features from French stop bursts

Deals with the perception of vowels from French stop bursts. The corpus was made up of 90 stimuli of 20-25 ms duration extracted from natural CVC (consonant-vowel-consonant) and CV words. The syllables combined the initial stops /p,t,k/ with the vowels /i,a,u/. In order to cut off all traces of vocalic segment, bursts whose duration was too short were lengthened. Eight native speakers of French served as listeners in the experiment. The results showed that a burst onset which did not contain any traces of vocalic segment provided substantial vocalic information (the overall identification rate was 80%). The vowel /i/ was clearly identified from /t/ and /k/, and the vowel /u/ very clearly identified from /k/. The vowel /a/, with high identification rate, was often chosen in the absence of a clear vocalic timbre.

Evaluation of phonetic constraints in acoustic‐to‐articulatory inversion

The objective is to evaluate phonetic constraints intended to guarantee the phonetic realism of vocal tract shapes generated by an articulatory model, i.e. here that of Maeda. The constraints are based on the relation between vocal tract shapes allowed for a vowel and formants frequencies of the vowel. They are speaker independent since they are derived from standard phonetic knowledge. We used them to reduce the number of solutions in acoustic‐to‐articulatory inversion. The impact of constraints has been investigated by comparing vocal tract shapes recovered by inverting five French vowels /i, e, a, u, y/ to those obtained by outlining articulator contours in X‐ray images. To that purpose we used X‐ray images (used by Maeda to elaborate his articulatory model) with the corresponding speech signal. The evaluation shows that unrealistic vocal tract shapes are penalized by constraints, that vocal tract shapes recovered through inversion are in very good agreement with X‐ray images and that relevant articula...

Using decision trees to construct optimal acoustic cues

This paper presents an approach to the optimization of acoustic cues used for stop identification in the context of an acoustic-phonetic decoding system which uses automatic acoustic event extractors (a formant tracking algorithm and a burst analyzer). The acoustic cues have been designed on the basis of acoustic studies on stops and spectrogram reading experiments. This ensures that these cues have a certain amount of discriminating power but we do not know either the optimal thresholds nor which combination of cues, are the most efficient. Therefore, we propose to use the decision tree theory to choose the most discriminating power and to improve their discrimination power. Considering the stop occurrences of a training corpus, the best cues are those which allow the decision tree leading to the best partition to be constructed. We have considered all the cues derived from the ones provided by the phonetician on formant transitions and burst characteristics. The improvement of the cues has been achieved on a corpus of 941 stops.