Boundary Tones Research Articles

Synthesis of falling nuclear pitch accents

A falling (“declarative”) nuclear pitch accent has a simple shape: an upward pitch obtrusion on the accented syllable, a fairly abrupt fall to near the bottom of the speaker's range, and a more gradual fall to the end. The synthesis algorithm in Anderson, Pierrehumbert, and Liberman (1983) decomposes this pattern into a H* pitch accent followed by an L phrase accent and an L boundary tone. It devotes four parameters to the realization of the H* accent (amount of rise, target F0 height, target duration, target time) and two parameters to the L phrase accent (target F0, target time). The parameters depend in part on properties of the accented syllable (e.g., prominence) and in part on characteristics of larger phrases (e.g., pitch range). The resulting time/F0 pairs are subject to interpolation and smoothing. This paper improves on the naturalness of the algorithm by incorporating findings from a recent study by Steele. In particular, the principles that determine the timing, magnitude, and alignment of the H* F0 rise and of the H* to L F0 fall have been improved.

On the interaction of tone and intonation in Mandarin Chinese

Intonation in many languages, including English, is realized most effectively by changes in fundamental frequency. It is important to investigate whether this relation holds true in a tone language, where F0 variation may affect lexical meaning. This paper reports the interaction of tones and intonation in a tone language, Mandarin Chinese, and calculates the effects of various intonation patterns as well as focus structures. The result is implemented and tested in a Chinese text‐to‐speech system. In general, the interrogative pattern in Mandarin Chinese is characterized by a high reference line and a high boundary tone at the end of a sentence. However, the resulting tone shape is not always rising as a result of the lexical tones. Focus is realized mainly by increasing the pitch range, but the exact implementation is complicated by some local tonal effects such as target shift. This paper will also explain the interaction of catathesis (pitch lowering triggered by specific tonal context) and focus in the direction of prosodic structure.

PROPERTIES OF I-BOUNDARY LENGTHENING OF VOWELS IN ENGLISH AND SERBIAN

Phonological features which signal the right IP edge include boundary tones and pre-boundary lengthening. However, languages differ in the prominence of these features. While in English final information focus has significant F0 prominence on PWd immediately preceding I-boundary, Serbian primarily relies on durational cues to signal final information focus and, thus, the right IP edge. This research aims at analyzing the degree and scope of I-boundary lengthening in English and Serbian as well as examining whether there is a compensatory relation between F0 prominence and pre-boundary lengthening in these languages. The research is part of a larger-scale corpus study involving approximately one hour of read speech per language. Both English and Serbian speakers were professional newsreaders and the recordings were comparable in terms of articulation rate and recording quality. The acoustic analysis involved measuring the degree and scope of lengthening in trisyllabic PWds, which was done in the Praat programme (version 6.0.52). The results of the measurements were statistically analyzed. The results of the analysis primarily point to the differences in the scope of I-boundary lengthening in English and Serbian. While the scope of lengthening in Serbian affects the stressed vowel of IP-final PWd as well as all the vowels following it, English post-accented non-final unstressed vowels are skipped by this process. Accordingly, wider scope of lengthening might suggest slightly higher significance of this process in Serbian, which can further be interpreted as a compensation for low F0 prominence at the right IP edge in this language.

An electronic acoustic image converter: Goldman, R. G.J. Acous. Soc. Amer., 34, p. 514 (1962)

A novel, free from paraxial approximation and computationally efficient numerical algorithm capable of predicting 4D acoustic fields in lossy and nonlinear media from arbitrary shaped sources (relevant to probes used in medical ultrasonic imaging and therapeutic systems) is described. The new WE (wave envelopes) approach to nonlinear propagation modeling is based on the solution of the second order nonlinear differential wave equation reported in [J. Wójcik, J. Acoust. Soc. Am. 104 (1998) 2654–2663; V.P. Kuznetsov, Akust. Zh. 16 (1970) 548–553]. An incremental stepping scheme allows for forward wave propagation. The operator-splitting method accounts independently for the effects of full diffraction, absorption and nonlinear interactions of harmonics. The WE method represents the propagating pulsed acoustic wave as a superposition of wavelet-like sinusoidal pulses with carrier frequencies being the harmonics of the boundary tone burst disturbance. The model is valid for lossy media, arbitrarily shaped plane and focused sources, accounts for the effects of diffraction and can be applied to continuous as well as to pulsed waves. Depending on the source geometry, level of nonlinearity and frequency bandwidth, in comparison with the conventional approach the Time-Averaged Wave Envelopes (TAWE) method shortens computational time of the full 4D nonlinear field calculation by at least an order of magnitude; thus, predictions of nonlinear beam propagation from complex sources (such as phased arrays) can be available within 30–60 min using only a standard PC. The approximate ratio between the computational time costs obtained by using the TAWE method and the conventional approach in calculations of the nonlinear interactions is proportional to 1/N∼2, and in memory consumption to 1/N∼ where N∼ is the average bandwidth of the individual wavelets. Numerical computations comparing the spatial field distributions obtained by using both the TAWE method and the conventional approach (based on a Fourier series representation of the propagating wave) are given for circular source geometry, which represents the most challenging case from the computational time point of view. For two cases, short (2 cycle) and long (8 cycle) 2 MHz bursts, the computational times were 10 min and 15 min versus 2 h and 8 h for the TAWE method versus the conventional method, respectively.