Speech bandwidth compression systems

摘要

936,124. Vocoders. MINISTER OF AVIATION. Dec. 18, 1961 [Dec. 30, 1960], No. 44730/60. Class 40 (4). In a formant tracker for an analysis synthesis telephone transmission system an adjustable network operates on the incoming speech and a signal is derived which is rectified and fed back to adjust the network continuously so that the network output is the original speech operated on by substantially the inverse of at least some of the formant forming filters of the vocal tract, whereby the signal derived is constrained to be representative of the corresponding formant parameter of the input speech signal. Fig. 1 shows a block diagram of a preferred embodiment for deriving tracking signals at O1 and 02, corresponding to the two lower formants of the speech input at I1, the tracking signals being amplified and fed back via I2 and I3 to control the characteristics of the network. It is shown mathematically that, assuming the speech signal can be represented as where p is the Heaviside operator, w 1 and w 2 are the two lower formant angular frequencies, and P 1 (t) is a function dependent on the excitation and other characteristics of the vocal tract, the output of adder S1, Fig. 1, will be (1+ap2)S(t), where a is the signal applied via input 12 to multiplier M1, (if a = 1/x 1 2 this x 1 2 function (1+ap2) is the inverse of the function which represents the effect of the formant angular frequency w 1 on the original speech and thus (1+ap2)S(t) will represent the speech signal with the component due to the formant w 1 removed. Similarly, the output of adder S2[= (1+ap2)] will represent the speech signal with the component due to formant w 2 removed. The output of adder S3 will be (1+abp4+ap2+ bp2)S(t) = (1+ap2)(1+bp2)S(t) and if a and b are correct (i.e. = 1/w 1 2 and 1/w 2 2 respectively) will contain no components due to either formants w1 1 or w 2 , but if one of the formants say w 1 , is incorrectly tracked there will be an output from this adder S3 which will be in phase with the output from one of the adders, i.e. S2, and these signals applied to the inputs of phase-sensitive detector P2 will produce an output dependent in magnitude and direction on the difference 1 between a and +- which output, after integraw 1 2 tion and amplification, is applied to input I2 to adjust the network to eliminate the formant corresponding to the angular frequency w 1 from the outputs of adders S3 and S1. A circuit is described with respect to Fig. 2 (not shown), incorporating compensation for the limitation of frequency response of practical differentiating networks.