Some physiological correlates
Let us arbitrarily characterize the activation phase as a sudden peak and the execution phase as a gradual slope as in the functions shown in Fig. 3.18, Curve I (the exact shapes are immaterial for the argument); I(a) shows the time of speech production and (b), (c), and (d) show the artificially delayed arrival times of the speech sounds at the ear. In I(b) the feedback is delayed by about 40 msec and in I(d) by about I65 msec. During the activation phase, the most rapid readjustments in the geometrical spaces of the vocal tract can be made, roughly corresponding to the times in which consonants are produced and/or lips and tongue are set in preparation of vowels; the execution phase is then primarily taken up by the vowel sounds themselves. Thus, the acoustic signals of the activation phase are rapid transitions in the speech wave. When these signals are perceived aurally, they may be assumed to serve as cues for the initiation of the next cycle.
讓我們任意地將活化期描述成驟起高峰，執行期為緩波形，可參看圖表3.18曲線圖I（精確形狀於此論點不是重點），I(a)顯示話語產生的時間，而(b)、(c)、(d)表示語音傳到耳朵的人為延遲時間。在I(b)反饋時間大約延遲40毫秒，在I(d)大約遲了165毫秒。活性期間，口腔內幾何空間可做最快速的調整。大約和發子音時或發母音時唇與舌頭擺放準備時間相符。執行期起初只接收到母音。因此，活性期時聽覺信號在口音波形中快速地轉變，這些信號聽覺上接收時，推估可能為下一次循環開始的信號。
By introducing delays into feedback loop, the cues for the beginning of the next cycle are delayed by an equal amount of time, thereby prolonging the execution phase. The behavioral correlate of this is the drawing out of vowels; this is the most characteristic aspect of speech under delayed auditory feedback. As the delay is increased, the duration of the cycle also is increased accordingly until the delay becomes so long that the signal of the first activation phase arrives at the ears at the same time as the motor apparatus is ready to produce the signal of the second activation phase. At this point, the activation phase of production coincides with the auditorily perceived activation signal, and the signal may now serve as the customary cue resulting in a sudden decrease in speech interference.
將這些延遲時間納入反饋循環，則下次循環的第一個信號會延遲相同時間，因此也延長了執行期的時間。說話時聽覺回饋延遲最典型方面為延長母音時引發的連帶關係。一旦延遲時間增加，循環週期也隨之延長，直到延遲時間不斷拉長，第一活性期的信號傳到耳朵，同時運動神經裝置開始運作第二活性期的信號，就這點來看，活性期的信息產生與聽覺接收活性信號一致，現在這種信號視為快速減少話語干擾的慣性信號。
FIG. 3.18. I. Hypothetical changes in the readiness for adjusting articulatiory organs. II. Speech interference waxes and wanes as a funtion of auditory delay time. (For explanation see text.)
Curve II of Fig. 3.18 shows the hypothetical waxing and waning of interference as a function of delay time. If speech worked with mechanical precision, we would perhaps be able to demonstrate a periodic rise and fall, as in an autocorrelation function. However, imprecisions in the mechanism, factors of attention and concentration, asymmetries in the syllabic structure of speech, and other factors seem to make it difficult to demonstrate more than a single peak in this function. Perhaps with better instrumentation the postulated periodicity may be brought out. Because of mainly statistical artifacts concomitant with pooling and averaging of data, the shape of the function II(a) will become less well-defined, resulting in a curve approximating that shown in II(b). Presumably, Fig. 3.17 represents the first peak of this curve.
圖表3.18的曲線圖顯示延遲時間增長和干擾減少的假設。如果說話產生和精密結構一同配合，我們也許可應證週期的升降，如同一種自動相互關係的作用。然而，結構的不精確性，專心、專注的因素，說話時音節不對稱性和其他種種因素似乎更難以證明超過單一以上高峰形式。也許會有更好應證週期的儀器能推算出，不過因為主要人們產生的數據和總平均數據相符，而II(b)大略地畫出曲線，II(a)也較無完整限定形狀，想必圖表3.17代表的就是曲線的第一高峰。
(b) Signal-switching Between Right and Left Ear. Cherry and Taylor (1954) have reported an experiment in which subjects were required to repeat instantaneously the sentences that were transmitted to them through the ear phones. This task is often called “shadowing.” Subjects can changes or distortions. Cherry and Taylor were interested in seeing how fast a subject could direct his attention from the input to the right ear that of the left ear. In order to answer this question they switched the speech signal alternatively from one ear to other, so that at any onetime only one ear could listen. Their device made it possible to vary the switching rate. In plotting switching rate against articulation score (Fig. 3.19) they found, rather surprisingly, that there is a “critical” switching rate at which articulation scores are lowest. That rate is about three switching cycles per second (that is, when each ear is allowed to hear one-sixth of a second at a time). Huggins (1964) duplicated Cherry and Taylor in an effort toward discovering whether the lowering of the articulation score is due to a disturbance in the perceptual process or whether it is due to some mutilation of the speech signal; the result of his work favors the latter alternative, although some switching of attention does seem to be taking place. His subjects varied a great deal in the amount of disturbance they experienced from the switching, and there were few with articulation scores as dramatically lowered as in the subjects reported by Cherry and Taylor. This particular difference between Huggins’ results and those obtained by Cherry and Taylor may be due to a number of factors that were not controlled in either of these studies; these factors need not concern us here. Huggins did confirm, however, that there is a critical switching rate, and this rate was constant for all subjects and the same as the one reported by the first authors. When the switching becomes too fast, subjects tend to concentrate on the input of their dominant ear, and therefore the switching, at certain critical rates, may have the effect of essentially interrupting the signal periodically. Nevertheless, Huggins’ subjects obtained greater accuracy in their shadowing task when the signal was switched from ear to ear than when presented with nothing but the interrupted speech sound of one ear. These are merely technical details. Cherry and Taylor, as well as Huggins, were confronted with the magical one-sixth of a second as a basic time unit in speech production. The connection is further clarified by the next point.
FIG. 3.19. Articulation score for continuous speech, switched periodically at various frequencies from one ear to ther other in the subject. For each ear, the proportion of period occupied by speech is 50%, the remaninder being silence.
(From Cherry and Taylor, 1954.)
(c) Rate of Interruptions. Periodically interrupted speech is unintelligible if the interruptions occur with certain frequency. Intelligibility is lowest at a rate of 3±1.5 interruptions per second, that is, when listeners are allowed to hear about 165 msec of speech at a time (with a lower limit of 110 msec and an upper limit of 300 msec). If the interruption rate is outside these limits, intelligibility is much improved, as shown in Fig. 3.20. These curves differ from the results of somewhat similar experiments reported by Miller and Licklider (1950), but the discrepancy must be due to the difference in stimulus materials used.
