Quick and Dirty Vocoder?

[frequencycentral]

Maybe over simplifying I know, but a vocoder is just a bunch of bandpass filters, VCA's, envelope followers and a mixer. You can use a quad opamp to make two bandpass filters and an envelope follower, add half a LM13700 and you have a VCA too. So four TL084's and two LM13700's would give you four channels. Ok, you have to add some more circuitry for an input buffer, mic preamp and a mixer. It may not be a EMS Vocoder 5000, but surely it would do something cool sounding?

Using the search function, the forum is littered with 'maybe not' posts by the illustrious Mr Hammer (which I respect) but I'd like to know exactly where I'm over simplifying. Does the bandpass of a Dr Q filter have too broad a 'pass' to be useable, does the filter really need to be more tightly focused?

Just throwing this out there.................

[stm]

I'd say that 8 bands are necessary if you want some degree of intelligibility.  A typical "voice band" to consider would extend from 300 Hz to 3400 Hz.  You may consider using 1/2 octave bandpass filters, starting at 300 Hz, then the central frequencies would be: 300, 420, 600, 840, 1200, 1680, 2400 and 3360 Hz.  Each filter should have a Q of 1.4 to comply with the 1/2 octave bandwidth requirement.

If you want to experiment with just four bands, I'd suggest you implement the bands at 420, 840, 1680 and 3360.  This allows you to add later the rest of the bands with minimum changes.  If using just four bands, each bandpass Q should be set for one octave, i.e. Q=0.7 (theoretically), however since the bands are wider than usual it might be wiser to narrow the bandwidth, perhaps with a Q=1, so as to improve selectivity a bit.

It takes a lot of components to have a functional vocoder, however as you imply in your post, it can be subdivided in smaller chunks a la divide and conquer.

Another aspect to consider, is that your instrument signal should have lots of harmonics so the voice spectra can be imprinted on top, so when using a guitar you should consider an overdrive to increase the harmonic content.

Take a look here for a general block diagram:  http://www.paia.com/ProdArticles/vocodwrk.htm

[rackham]

Good thinking

I've been pondering the same thing since you mentioned the 'Iron Lung' recently.

That apparently has 256 bands so I'm thinking 4 bands will either be super-subtle or super grungy. If it's the latter, then ace!

Paia do an 8-band vocoder that's marketed as being pretty decent (I just checked out the audio demos, it sounds great), so maybe 4 might well sound vocoder-ish? Either way, I reckon picking your centre-frequencies for the filters is going to be the key

Presumably, if the Dr Q filter isn't narrow enough then you could up- the-Q to make it workable?

edit: Um, what STM said!

[Mark Hammer]

Vocoders are like EQ units.  If the EQ is centered on the "right" resonant bands, then you may only need 2 or 3 bands.  So why do they generally come as 7, 10, 15 or 31 band units?  Because you have no idea in advance what the "right" bands will be if you pack up your gear and play different rooms every night.

Similarly, it is likely possible to produce a vocoder that is nicely responsive to the particular voice input with less than 10 band, if you know the parameters of the specific input in advance and the filters are optimized for that voice.  The reason why so many vocoders have more than the 8 bands of something like the PAiA unit is because they are intended to be used with any voice you wish to feed it.  Could be a child's voice, a woman's, Barry White's, Donald Duck's, Lemmie's, they'll all "work" if you have a few dozen bands.

Now the second related question is what we mean by "work".  If the intent is to have articulate speech sounds then we need lotsa bands.  If the intent is simply to have an instrument signal tonally modulated in some fashion by voice, then yo can actually do with fewer.

Some years back, I proposed a talked-to pedal idea, in which voice would feed a handful of filters, each with their own envelope followers, and the individual envelopes could be used to control parameters in a different circuit.  They could be  filters, be they didn't need to be.  For example, lowering your voice might adjust tremolo rate, while shrieking in your highest voice cranks up the gain in a distortion stage, and so on.  The basic idea is that controlling circuits needn't be restricted to your hands and feet.  In a sense, it is a sort of extension of breath controllers used in some synths.

[doc_drop]

I have found that being able to mix in some of the original sound is the easiest way to make vocoders intelligable. Otherwise it is a nightmare knowing what sound to use as the carrier, etc. The software vocoders I've used all had lots and lots of bands, but even then some original signal in the mix really helps to make the words understandable.

It might be worth having a mini mixer in the circuit for that purpose.

[frequencycentral]

I think most vocoders feature a highpass filter on the modulator signal to add sibilance to the final mix, as this makes it more intelligable.

[doc_drop]

That's right. Which makes sense since it is the ssss's and often t's and d's that get smeared through vocoders. So, I guess just adding back the highs would be a good way to do it.