ETI VOCODER September - October 1980, designed by Richard Becker of Powertran A vocoder could be simpiy defined as a device which will, in real time, superimpose the spectral character- istics of one signal upon another. To Ieave it at that, however, would result in many a yawn and a few skipped pages. In fact vocoders are anything but boring! Put speech and the output of an instrument into a vocoder and the instrument not the operator, appears to be doing the talking or singing! Use the internal excitation oscillators, change the frequency and the speaker suddenly changes sex. Use the noise generator and these is whispering in the breeze. Use the output of a cassette deck and the London Symphony Orchestra recites the Karma Sutra! Just a few of the possibilities! Hurnan speech is built up from two basic components - the sound from the vocal chords which buzz when air is passed over them and the sound of air rushing past the teeth. These sounds are used to produce voiced and unvoiced speech respectively. By opening and closing the mouth and the nasal cavity, and by moving the tongue, thereby adjusting the resonances, the basic sounds are rnodified in amplitude and harmonic content. if the variations in amplitude and harmonic content can be analysed and applied to suitable electronic control circuitry then the basic sounds of speech can be substituted for by almost anything and this is just what a vocoder does. The first part of a vocoder is a spectrum analyser producing control signals which are a measure of the strength of the speech signal in each of the frequency bands (14 in this design). The substitution (excitation) signal is also split into a number ot frequency bands (using identical filters to those used for analysis) and each of these signals is passed through a voltage controlled amplifier whose gain is deterrnined by the control signals. The sum of the outputs of these amplifiers is the vocoder output. The system The speech signal, after passing through the preamplifier and tone control stages, is separated into 14 bands by badpass filters, a low pass filter and a high pass filter. The bandpass filters are double tuned that is to say each of the two stages has a slightly different resonant frequency. The effect of this is to broaden the band of accepted frequencies and give the response curve a flattened top. A high Q rnakes the filters cut off rapidly out ot the pass bands. The envelope foIlowers consist of an active full wave rectifier and a low pass filter, the output of which is the controI signal for the synthesiser section. The control signal passes through a sample and hold stage which is used to freeze the music, by means of a footswitch, at any required point of articulation. The stage is also used for slewing rate control which smooths out the control signals for slower and smoother changes in spectral balance and amplitude resulting in speech being changed into singing or chanting. Holy Responses In the synthesis section there is a filter bank identical to that of the analysis section. Voltage controlled amplifiers modulate the outputs of these with the control signals from the analysis section. The outputs are then summed to produce the output signal. Alternate channel outputs are inverted since there is a change in phase as a signal is swept through the resonant frequency of the filter. Therefore, at the midpoint between adjacent bands phase cancellation will þccur producing deep holes in the overall frequency response. By having adjacent channels outputs inverted with respect to each other there is addition instead of subtraction at the midpoints. For external excitation, there is a pre-amplifier and tone control circuit similar to that used for speech. The output of this stage is mixed with the two oscillators (which generate pulses of variable width and frequency) and also with the output of the noise generator. The noise also passes through an AGC circuit to match its level to the excitation signals. This noise is then used to substitute for the other excitation signals by the voiced unvoiced detector electronic switch when unvoiced speech is detected by the comparator which determines whether the majority of the energy in the speech is at low frequencies (2 kHz - voiced) or at high frequencies (4 kHz - unvoiced). Start construction with the power supply PCB and bolt this onto the rear panel with mica washers between the panel and transistors, which are on the underside of the board. Silicone grease will keep the washers in place during fitting. Wire up and check all is well when operating into 1k0 resistors as temporary loads. Build up the rest of the boards. Use insulation on any links which touch the leads of components. On the LED PPM boards, fit the connector pins for the connector to the component side of the board for the excitation meter and the noncomponent side for the speech meter. Where there are jack sockets, solder short lengths of bare wire to the boards and fit both board and socket to the front panel before soldering the wires to the sockets, which then become firmly attached parts of the board assemblies. On the internal excitation board of the three controls with mounting frames fit to the underside of the board. The other four controls fit on the top side of the board. To get the correct spacing between the top and bottom controls the top ones are soldered to pins such that the tags just touch the top of the board. Split Supply The analysis/synthesis board is split in two halves to simplify manufacture. When the two halves are completed,fit the boards to the panel by means of the controls. Fit wire links between the boards and solder the two halves together by use of a bared length of wire along the joint. Fit the spacers to the back of the board and drop it into the chassis so that it can sit on its back edge‚ whilst wiring up and setting up. Link together with a stretched length of wire on the back of the board all the excitation input pins and also the other inputs and outputs (a total of seven rails). Make up the wiring loom and connect up all the boards. Setting Up Check the power lines are still correct when all the boards are connected, set all presets to the centre of their travel and apply a sinusoidal signal to the speech line input. (If no oscillator is availabie use the cheap little circuit shown.) Set the level to where the sixth LED up just flickers, corresponding to 400 mV. Measure the AC voltage on pin 6 of IC1 channel 2. Adjust the frequency until the voltage reaches a peak, turn PR1 fully clockwise and turn it back slowly until 4 V RMS is measured at the resonant peak of the filter. Repeat this for the other analysis filters. Connect a 56R resistor between the bias rail and + 12 V, turn the slewing rate control fully clockwise check the pulse generator is operating by listening for a whistle when the input jack of an amplifier is placed near the slew rate control board, switch off the unvoiced detector, plug the oscillator into the extemal excitation HIGH input and set up channel 2 excitation filter as for the analysis filters (except that now the point to measure is pin 1 of IC7 and the potentiometer to adjust is PR5). With RV1 fully clockwise adjust PR2 so that 4 V RMS is also at the output of the OTA buffer (IC6 pin 8). Repeat this for the other filters including channels 1 , 14 where there is only PR2 to adjust Plug the vocoder into an amplifier, turn up all channel volume controls and the vocoder output control. Turn down the speech and the excitation inputs. Turn up one of the oscillators and adjust RV5 or RV6, as appropriate, so that the signal is heard to just disappear when the width control is anticlockwise. Repeat for the other oscillator. Noise Abatement Remove the 56R resistor turn down all the channel volume controls and the oscillators. Tum up the noise level to maximum, turn up channel 1 and adjust PR3 to the point just before the noise disappears. Repeat this for the other channels. Disconnect the excitation and speech inputs from the analysis/synthesis board and temporarily connect the excitation to the speech input of the board so that noise can be applied to the analysis section. Turn up channel 1 and the noise control. Adjust PR4 for minimum breakthrough of the control signal which will be heard as a low rumble. Repeat this for the other channels and then re-connect the inputs. Turn on the voiced/unvoiced detector, apply a high frequency signal to the speech input and the V/UV LE D will light up. Turn up all the channel volume controls and the noise control and noise will be heard. Adjust PR2 to halfway between the point where the noise is heard to start limiting. Turn down the noise control and adjust PR1 to the point where noise is just heard to disappear. Turn up the noise again and alter the frequency of the test oscillator..Adjust PR3 to where the noise level drops by about 6 dB, as indicated on the LED PPM, when theV/UV LED is illuminated.