| 摘要 |
1,113,224. Vocoder pitch extractor. MINISTER OF TECHNOLOGY. 14 Oct., 1965 [16 Oct., 1964], No. 42320/64. Heading H4R [Also in Division G1 In a larynx excitation period detector the speech signal is divided into a number of frequency bands and the time elapsing between successive major peaks of the signals in each of the bands is measured and the results of these independent measurements are correlated to obtain a measurement based on the best correlation. Fig. 1 shows one embodiment in which the incoming speech is passed through a lowpass filter 1 and modulated in a double balanced modulator 3 with a 10 kc/s. carrier from an oscillator 2. The resulting signal, in which the modulation to 10 kc/s. makes the envelope of the speech signal more readily distinguishable, is applied to a number of period measuring channels 5 to 9 which include band-pass filters 11 to 15 tuned to different frequency bands to enable excitation function period measurements to be made at a number of places in the frequency spectrum. The filters 11 to 14 are each 500 c/s. band width and cover the range 1050 to 11,300 c/s., each filter overlapping the adjacent filters by 250 c/s., while filter 15 covers the range 9500 to 10,500 o/s., i.e. the fundamental frequency range of the original speech, and the output from this filter, after passing through a logarithmic amplifier 25, is fed into two double balanced demodulators 26 and 27 fed with the 10 kc/s. carrier which is phase shifted in network 4, to compensate for the phase shift in filter 15 &c., and applied in opposite phase in the two demodulators so that the period measuring circuits 35 and 36 operate on the positive and negative peaks respectively in the speech waveform. Each channel contains a logarithmic amplifier 21 to 25 which increases the dynamic range of operation of the period measuring circuits. Each period measuring circuit 31 to 36 detects peaks in the waveform supplied to it and produces an analogue voltage the value of which is dependent on the time between successive major peaks in the waveform. The analogue peak period voltage is then applied to a comparator circuit, which may be of the type disclosed in Fig. 4 of Specification 1,113,222, which has applied to it a staircase voltage from the waveform generator 61. The comparator then produces a pulse whose position in time, relative to the reset time of the staircase voltage, is dependent on the period measured by the respective p.m.c. 31 to 36. By this means the outputs of the channels 5 to 9 are all referred to a common time frame. The outputs from the comparators 41 to 46 are all added at the input to the gate 49 so that pulses from the different comparators at the same time, representing similar measurements of the period in the different channels, reinforce to produce a higher pulse on the input of gate 49 at a time of occurrence corresponding to the period meassured. The gate is enabled by the 12À8 kc/s. clock pulse so that samples of the gate input are passed to the main peak detector 50 at a frequency of 12À8 kc/s. a further input to gate 49 from NOR gate 62 inhibits transmission through the gate for the two sampling instants at the start of each staircase comparator waveform to ensure that the digital values of period 000000 and 000001 are not transmitted. The binary counter including stages 53 to 60 is driven from the 12À8 kc/s. oscillator 51 and provides the drive for the staircase generator 61 so that the count recorded in stages 53 to 58 indicates progress up the staircase or a digital indication of the time elapsed since the start of the staircase waveform. The main peak detector 50 produces an output whenever the signal input increases above a reference level established by the maximum voltage attained by the signal input in the period since a reset pulse was applied to the main peak detector. The output of the peak detector 50 is applied via the " OR " gate 76 to the gates 68 to 73 so that the count in the bi-stable stages 53 to 58 is entered into store S0 The count in store S0 therefore indicates the time at which main peak occurred. The two bi-stable stages 59 and 60 provide, via gate 74, a sampling pulse once every four cycles of the staircase waveform and this pulse, applied to gate 77, causes the transfer of the count in store S0 into store S1 and also, via gate 75, causes the main peak detector 50 to be reset. Thus the position of the maximum value o the waveform at the input of gate 49 which is attained during 4 cycles of the staircase waveform, is registered in store S1 to be used as a digital indication of the period of the input signal. In an alternative arrangement, Fig. 9 (not shown), for use when the speech signal is derived from a high-quality microphone which transmits the low frequencies in the speech, fewer period measuring channels are used and the double measuring channel (9) covers speech frequencies up to 250 c.p.s. only. In addition the signal in the store S0 is sampled at the end of each staircase cycle and this signal, after conversion to analogue form in a digital to analogue converter (128) is compared with the staircase waveform in a comparator (47) similar to those of the measuring channels (5 and 6) to provide an additional signal which tends to emphasize measurements of period which lie close to the value determined during the previous cycle of the staircase waveform. Period measuring circuits (31 to 36, Fig. 1) are shown in detail in Fig. 6 in which the selected frequency band of the speech signal is applied to line 89 and pulses at 12À8 kc/s. applied to line 90 so that they are at the base of Q1. Transistors Q1 and Q2 operate to shift the D.C. level of the speech signal so that only the positive peaks of the signal make transistor Q3 conduct so that only the pulses added to the waveform appear at the collector of Q3. These pulses cause, via transistors Q4 and Q5, capacitor C2 to be discharged to cut off transistor Q6. A predetermined time after the last pulse Q6 conducts to cause Q7 to conduct so that the collector waveform of Q7 is a rectangular wave whose period is the period of the peaks in the incoming speech waveform. The rectangular wave is differentiated and applied so that the negative-going edges fire monostable trigger circuit T1 the return of which fires trigger circuit T2 which discharges C3 which thereafter charges logarithmically until T1 is triggered again which causes transistor Q17 to conduct and copy the potential then existing on capacitor C3 on to capacitor C4, which potential is applied via emitter followers Q18 to Q20 as the period measuring circuit output. The output of the emitter follower Q18 to Q20 is also applied via Q21 to a comparison circuit 95 where it provides a reference voltage to be compared with the voltage on the collector of Q17 which is the voltage on capacitor C3. The comparison circuit provides a gating signal for the pulse input to transistor Q8 and prevents the firing of trigger circuit T1 until the voltage on C3 is within a predetermined voltage of that on the output line 94, and thus prevents the transfer of a voltage to C4 which is grossly smaller than the voltage transferred on the previous measurement. Main peak detector, Fig. 7, is used to determine the maximum positive excursion of the signal voltage applied to line 48 during a predetermined time interval and it comprises a capacitor C5 which at the beginning of the time interval is clamped to a predetermined voltage by a reset pulse applied to line 96, the clamping voltage being set by the potentiometer P. Whenever the input signal voltage rises above the voltage on capacitor C5 the capacitor is charged through transistor Q25 to the voltage on line 48 and this voltage is maintained across C5 until either the input signal voltage exceeds the new voltage on C5 or a reset pulse is applied to line 96. Transistors Q29 to Q32 provide a clamping action on the input line 48 so that the circuit is inoperative for the duration of the signal applied to line 98 and so that the signal fed to Q25 is in the form of samples taken at the frequency of the pulses fed to line 99, i.e. 12À8 kc/s. A complete vocoder system incorporating apparatus according to this invention and also including an error discarding apparatus according to Specification 1,113,222 and interpolating and extrapolating apparatus according to Specification 1,113,223 and formant tracking apparatus according to Specification 1,113,221 and an unvoiced sounds detector according to Specification 1,113,225 is described with reference to Fig. 8 (not shown). An alternative vocoder system. Fig. 10, incorporates a smoothing filter 138 for the excitation function signal which is switchable between a long and a short smoothing time constant by a bi-stable unit 141. When voiced sounds occur the binary signals representing them are fed via store 130 and gates 135, operated by a timing signal fed via a gate 133, to a store 136 and a digital to analogue converter 137, the analogue output from which is fed through filter 138, set to its long time constant condition, to the larynx tone generator 109. When an unreliable measurement occurs, i.e. indicated by a 000001 digital signal, NOR gate 131 is operated which inhibits gate 133, preventing the timing signal enabling gates 135, so that the last reliable measurement transferred to store 136 is retained to be used in place of the unreliable measurement. When an unvoiced sound is received, indicated by a 000000 signal, adder 132 produces an output, and the leading edge of this pulse operates " beginning " unit 140 to operate bi-stable 141 to change the filter time constant to its " short " condition so that it may respond more quickly to a change in pitch on the next reliable measurement. The next reliable measurement of excitation period in store 130 causes the cessation of output from " NOR " gate 131 which is detected in " end " element 139 and |
