| 摘要 |
1,233,462. Vocoders. INTERNATIONAL BUSINESS MACHINES CORP. 4 June, 1968 [14 June, 1967; 20 March, 19681, No. 26535/68. Heading H4R. In a vocoder the speech signal samples are compressed in time and applied repetitively to a single spectrum analysis channel, the signal being shifted in frequency by a modulation system between each application to the spectrum analysis channel so that the sequential output of the single channel represents the spectrum function of the speech signal. At the receiver synthesizer the incoming signals are reconstituted as high frequency time compressed channel signals, using a stepped oscillator and a single synthesizer filter, and the compressed channel signals are added and expanded in time to produce the synthesized speech. Spectrum function analysis, Fig. 5.-The speech input at 40 is converted to digital form in delta modulator 41 and the resulting digits, at 20À1 Ás intervals, are fed, in groups of five, into register 53 via gates 43 to 47 opened sequentially by ring counter 80. Digits from register 53 are read out periodically at high speed, under the control of counter 92 and gates 54 to 58, into a delay line 63. The signals in the delay line are continually circulated and the capacity is such that a total of 1000 pulses are stored therein at 0.1 Ás intervals, so that the delay holds the last 20.1 MS. of speech compressed in time to occupy only 100 Ás, i.e. compressed by a factor of 201. The digital signal from the delay line is applied via a delta demodulator 66 and low pass filter 67 to a modulator 68, the signal now having a frequency range 60,300-804,000 Hz., instead of 300- 4000 Hz., due to its compression in time. The other input to modulator 68 is derived from an oscillator 69 whose frequency is modulated in a stepping manner through 100 values between 1.26 MHz. and 2 MHz., the output changing every 200 ÁS. The resulting signal at the modulator output is applied to a band pass filter 70 of band width 20 KHz. and centre frequency 2.06 MHz., the arrangement of modulator 68 and band pass filter 70 simulating a speech spectrum analysis system including a bank of 100 band pass filters of band width 100 Hz. at 37 Hz. spacing of centre frequencies. The output of band pass filter 70 is applied to a level detector 71 which stores the maximum levels of the analysed signals. A base band section, e.g. 60.3 to 180.9 kHz., of the time compressed speech, corresponding to 300-900 Hz. in the original speech, may be filtered off at the output of low pass filter 67 to be used as an excitation function signal. Synthesizer, Fig. 2.-The excitation function signal, comprising time compressed repetitive versions of the base band of the speech signal, is applied, via line 22 to circuits 25 and 26 which, by distorting the base band excitation signal, provide a wide band signal covering a range equivalent to the full speech spectrum which is applied to a modulator 27, fed with a carrier from an oscillator 28 whose frequency is stepped synchronously with the oscillator in the analyser, the resulting signal is fed through a band pass filter 29, similar to the analysis filter, modulated by the spectrum function signal applied over line 23, in modulator 30 and restored to the appropriate frequency band by a second modulator 30A fed with the same carrier, from oscillator 28, as the modulator 27. Therefore on line 31 appears successively time compressed signals corresponding to the outputs of synthesis channels which are applied to the time expander 24. The successive signals are converted to digital form in modulator 33 and applied to delay line 34. The various responses as the synthesis filter is effectively swept across the speech frequency band are added together by recirculating the output of the delay line back and adding it to the input at 32, via an appropriate demodulator 38 if this is necessary according to the type of modulator used at 33. When the total response has been built up in the line the signal is sampled, in groups of five bits and extracted via buffer circuit 35 and serializer 37, the output pulse train, read out at low speed from serializer 37, is demodulated by 39 to give the synthesized speech output. If the excitation function is below a predetermined level at the input to the demodulator 27 it may be replaced by the output of a random noise generator, Fig. 9 (not shown). The analyser and synthesizer may both use the same delay line by interleaving the signals in the line, Fig. 13 (not shown). Also the delay line may be used in time division multiplex to handle the signals from a number of lines, Fig. 14 (not shown). |
