Analysis of airflow and air volume during continuous speech

Abstract

Program RESPAN is a respirometric analyzer operating on airflow (volume velocity) function during speech obtained through a pneumotachograph-pressure-transducer-amplifier system (Hardy, 1965; Isshiki & Ringel, 1964). The volume velocity function during speech typically consists of large negative portions corresponding to inhalation and slowly varying positive de signals (phonetic and syllabic components) superimposed by higher frequency ac components (voice fundamental frequency for voice segments). Specific functions of the program are (1) to detect inhalation and exhalation, (2) to calculate total inhaled and exhaled air volume for each breath group, (3) to obtain time ratio of contiguous inhalation and exhalation, (4) to generate piecewise linear approximation of the volume velocity function, and (5) to calculate the maximum, minimum, and average rate of air usage for closureto-closure segments (segments between a point of zero volume velocity to next zero volume velocity during speech) within each breath group. The program performs as a combination of peak detector, valley detector, plateau detector, zero detector, and inhalation detector. Primary programming consideration was given to reject the high-frequency (voice fundamental frequency and quantization noise) components in a peak-valley-plateau detection and to process multichannel signals of virtually unlimited length. The multichannel processing has been implemented into the program because of frequent needs to simultaneously investigate airflow, air volume, and voice signals. The present program accepts a five-channel data and allows a maximum signal length of about 1 h. Input/Output. The data collection procedures are illustrated in Figure 1. Channels 1 through 5 are used for voice signals, airflow signals with high-gain amplification, integrated airflow (cumulative volume) signals, the airflow signals with low-gain amplification, and identification signals (-10 V de), respectively. The purpose of the low-pass filtering for flow and volume signals is to minimize voice fundamental frequency components. Signals on the five channels arc scanned and quantized through an analog-to-digital converter at 5,000 times/sec, resulting in a I-KHz sampling rate for each channel. The digitized data are written on a computer magnetic tape through a chaining buffer. Use of the two levels of amplification for airflow signals is motivated for optimal use of dynamic range of the system for both relatively low exhalatory airflow during speech and considerably larger airflow during the inhalation. If a single channel were used to record both inhalation and exhalation, resolution of airflow signal during speech would be considerably reduced. Voice signals on the first channel arc used to temporally relate the aerodynamic events with acoustic and articulatory events. Since voice signals are picked up by a microphone inserted incident to air passage in the pneumotachograph, they have predominantly low-frequency components (due to the resonant characteristics of such a system). The purpose of the identification signal of -·10 V dc on the fifth channel is to allow the program to double-check and align, if necessary, the sequence of the channels for each scan and eliminate any extraneous data from analysis. The integration signal (reset either every second or every J0 sec) provides cumulative volume measures and is primarily uscd to compare and validate results of this computer analysis proccdure with those obtained by other techniques that employ such integration methods on volume measurements (Hardy & ,.~-----...... --_ ...._-.. : CDC 17~ : I COMPUER : I • . . I • , . I ! : I , L .... • ._.__ J