I've had a DIY Vocoder project in the back of my mind for much of the last six months. I dredge it up now and again, and look at it, see if my thinking regarding the difficulty, or my ability to afford it has changed and then put it back. Finally, I've had enough of this, and it's time to start putting the idea to paper with a view toward building the thing.

Homer Dudley's 1930 invention of the Vocoder (a portemento of the words Voice Coder), was an answer to the need to conserve bandwidth on heavily trafficked long-distance telephone lines. By only sending the slowly changing voice envelope information (as a series of vocal formants), and then re-synthesizing the speech on the other end using a noise source, intelligible, if somewhat robotic-sounding speech could be heard. The technology had applications in cryptography, allowing the transmission of secured voice messages (and indeed completely confounded the Germans in WWII).

Dudley's design was, of course, fully analog, involving filters, peak detection, amplifiers and mixers. The model for how vocoders in general work is still valid, even with the advent of DSP techniques like phase vocoding and Multiple Excitation Linear Prediction (MELP) vocoding.

I'm not going to try to draw a diagram just yet, but here's how it goes in terms of Dudley's filter version.

The modulation signal (can be and usually is, a human voice), is passed through a fixed filter bank, with each bandpass filter having it's own independent output channel. Each channel is then passed through its own peak detection or envelope follower circuit. This represents the analysis portion of the device.

The signals from the analysis portion are then passed to re-synthesis. The vocal formant data in the form of envelopes is used to control the gain on a bank of voltage-controlled amplifiers. A carrier signal is fed through a second fixed filter bank, whose outputs and sent to the amplifiers. Each channel is then input into a mixer, and out comes speech at the other end.

It was none other than Wendy Carlos, in the late 1960's having seen vocoding in action, who realized that the carrier signal didn't have to be a fixed noise source, but could, in fact, be a musical tone, which would result in a "singing synthesizer". Carlos was perhaps just naive or silly enough to ask Bob Moog if he could make "something like a vocoder", using the standard Moog modules of the time, which Bob being Bob, he did. The results of this can be heard in all their glory on Wendy Carlos' "Timesteps", which was used in part in Stanley Kubrick's "A Clockwork Orange", along with an except from Beethovan's Ninth, which now sounds rather dated, but nonetheless features among the first musical uses of the vocoder.

Moog's original kludge implementation lays out all the details of how a vocoder works by constructing it in modules, and thus, understanding it becomes a wonderful tool for understanding vocoders in general. In terms of putting together a DIY project, building it as discrete modules, even if they're then placed behind a single panel, makes excellent sense.

After Moog and Carlos, the Kraut-Rock phenomenon of Kraftwerk designed and built their own vocoder, and audio pioneer Harald Bode designed a 16-band unit which later was sold under the Moog label. The absolute pinnacle of musical vocoder design was arguably the EMS Vocoder 5000, but these were large and expensive studio units. Dutch manufacturer Synton made much more affordable (still pricey) units, with great performance and intelligibility.

Of course, one can easily achieve the same effect nowadays in software using DSP techniques, and any number of software vocoders exist. But to me, that's just no fun at all. There is also a nice 8-band vocoder kit from PAIA, but again, it's not floating my boat. Give me a good pair of fixed filter banks, some envelope followers, some VCAs and a mixer, and let me toss them together and see what I get.