JPS51135304A - Speech pattern takeout system - Google Patents

Abstract

1533112 Speech pattern derivation LARYNGOGRAPH Ltd 8 Jan 1976 [24 Jan 1975] 3269/75 Heading G1N(E) [Also in Division G5] In a laryngograph for monitoring the vocal fold vibrations in a speaker's larynx in speech training &c., a single frequency oscillator, Fig. 2, or a broad band oscillator, Fig. 3, supplies a carrier wave to a transmitting aerial positioned adjacent the larynx. The modulated carrier wave is picked up by a filter further adjacent aerial whose output is detected 3, Fig. 2, or demodulated 8, Fig. 3. The result is filtered 6 to produce a vocal fold waveform Lx representative of the "voiced" sounds in each utterance. The carrier wave level is maintained constant by two alternative types of feedback system controlling the oscillator 2 or the detection circuits (which may employ slicing techniques). The electrode aerials may take the form of a standard printed circuit board which is etched to produce a circular element with a concentric ring spaced therefrom. Larynx waveform L X displays and vocal fold vibration determination. The signal from the filter amplifier 6 is fed via a band pass filter 40, Fig. 4, which supplies an amplitude sensing circuit 51 connected to a memory input control 52. The control 52 controls the input pulses, from a clock 50, which are gated 48 under the control of a closure pulse generator 42, supplied, as shown, by an infinite clipping circuit, to provide a uniform pulse for each larynx closure. This enables pulses to be counted 49 to give a measure of the vocal fold vibration period Tx which is eventually stored in a main memory 54. The pulses from the counter 49 are directed by the control 52 into a random access memory 54 connected to a display unit 57. The signals reaching the display must be logarithmically converted at 56, or at any other convenient point in the system. Alternatively, suitable non- linear clock pulses may be used. A chart recorder 45 or a "X-Y" plotter may feed the memory via a buffer memory so that only voiced segments of the speech are drawn slowly, while intervening voiceless pauses are moved over quickly. In a further embodiment, Figs. 6 and 7 (not shown), the characteristic frequency in two frequency bands F 1 , F n are shown on a stored display with a colour distinction between parts of an utterance which are voiced, fricative (consonants) or voiced fricative in form.