| 摘要 |
1,192,215. Vocoders. INTERNATIONAL BUSINESS MACHINES CORP. 8 Aug., 1967 [16 Aug., 1966], No. 36326/67. Heading H4R. In a vocoder in which signals are derived representative of the energy in each of a number of frequency bands within the speech spectrum, and in which the signals are sampled sequentially and transmitted in time division multiplex, if a group of channels relating to the high frequency end of the speech spectrum do not carry significant amounts of energy, as may occur during a voiced sound, or a group of channels at the low frequency end of the speech spectrum do not contain significant amounts of energy, as may occur during an unvoiced sound, then signals relating to those channels are not transmitted, a signal being transmitted only to indicate which channels are omitted. Fig. 3 shows the analysis section of a vocoder SPA which produces signals or lines 1 to n representative of the energy distribution over the speech frequency spectrum. The signals on lines 1 to n/2, representing the lower frequency portion of the spectrum, are fed to a device FCD where signals above a predetermined threshold operate Schmitt triggers S, the outputs from which operate bi-stable switches T at predetermined times during the sampling cycle, which is arranged by the clock generator CL to be at the frequency of the speech excitation signal. The output from the switches T is fed via the system of "OR" gates OG to derive a signal on the lines 1 to n/2, feeding digital to analogue converter D/A, consisting of a number of zeros equal to the number of successive vocoder spectrum channels, starting from the lowest frequency channel, having energy levels below a predetermined threshold. The converter D/A provides an analogue output having a value proportional to the number of leading zeros in its input signal. The analogue signal is supplied as an input a1 to the time division sampling switch SW while the input signal to the converter D/A is also fed to a logic circuit LOGF to operate one of the gates AG1 to n/2 + 1 so that the time division sampling pulse from the clock generator CL via delay circuit #Ta, is fed into the appropriate stage of the shift register, comprising delay stages #T1 to #Tn, so that successive channels relating to the lower frequency end of the speech spectrum which contain less than the predetermined threshold energy value are not sampled. In a similar manner the energy levels in the channels relating to the upper end of the speech spectrum are monitored in the device LCD and signals from consecutive channels, at the upper frequency end of the spectrum which contain insignificant energy are not transmitted. The resulting signal fed on the line M-PAM-O therefore comprises first a pulse whose amplitude is dependent on the number of channels at the bottom edge of the spectrum which contain no energy, then a pulse whose amplitude is dependent on the number of channels at the top edge of the spectrum which contain no energy, and finally amplitude modulated pulses representing the energy in successive remaining channels, in order from low frequency to high frequency. Fig. 4 shows a receiver in which the incoming time divided pulse stream is fed to synchronize a clock pulse generator CL1, the output from which is fed via a shaping circuit SH to a shift register of delay stages #Ta, #T1, - - - #Tn, to provide gating signals for the commutating switches SW. The first amplitude modulated pulse to arrive is fed via a switch SW to the analogue to digital circuit FCC so as to cut out of circuit those shift register stages relating to switches appropriate to channels at the lower edge of the spectrum which have been omitted from the transmission as having negligible energy content, similarly the second amplitude modulated pulse is fed to the circuit LCC to do the same for the upper edge of the frequency spectrum. The remaining pulses are fed through their respective switches SW to the synthesis filters BPF 1- - - n to provide the synthesized speech at terminal SPO. |
