Question about the HVM

[kdjupdal]

The Human Voice Module is a fixed filter bank that uses seven fixed bandpass filters. Can you get the same effect by using seven bandpass filter modules (like the COTA or SVF)? Probably 4 would be enough, but, are the filters similar or are the filters in HVM of another type, for instance steeper slope?

[Synthbuilder]

Can you get the same effect by using seven bandpass filter modules (like the COTA or SVF)?

Not quite the same but similar. Four standard band pass filters should be sufficient for a close enough approximation. I've made versions of it using the BP response in plugin EQs.

Note that the HVM, which itself is a copy of the filter bank in a VP-330, is made from a parallel network of dual Deliyannis BP filters. In each band there are two BP filters in series. One filter in each pair has a slightly different centre frequency to the other, simply be having different resistor values. This gives a wider band than would otherwise be obtained from a single filter.

Tony

[kdjupdal]

...
Note that the HVM, which itself is a copy of the filter bank in a VP-330, is made from a parallel network of dual Deliyannis BP filters. In each band there are two BP filters in series. One filter in each pair has a slightly different centre frequency to the other, simply be having different resistor values. This gives a wider band than would otherwise be obtained from a single filter.

Two bandpass filters in series, wouldn't that mean that we get 12db/oct slope instead of the normal 6db/oct slope that bandpass filters normally have?

[Synthbuilder]

Two bandpass filters in series, wouldn't that mean that we get 12db/oct slope instead of the normal 6db/oct slope that bandpass filters normally have?

Band pass filters can come in a different slopes. The traditional SVF band pass output is indeed -6dB/octave, but other filter topologies will allow for steeper slopes.

However, slope is not always that helpful when it comes to trying to understand what a filter will sound like. The -24dB/octave slope that four pole filters are quoted as having is true, but only at frequencies well beyond the cut-off frequency, and by then the gain of the filter is quite low. Quite often the Q factor or resonance will have a greater impact on the sound of the filter.

A two stage deliyannis filter with identical centre frequencies will produce eventual 12dB/octave slopes but only at some distance away from the centre frequency. Nearer the centre frequency the slopes will be steeper because of the high Q of the filters. Two slightly offset Deliyannis filters will also eventually tend to -12dB/octave but it's the way they interact nearer their differing cut-off points that will affect the sound far more. The frequency response looks a little like two mountain peaks merged together to form a single double topped mountain with very steep sides.

The steeper slopes do produce better vocal sounds, but good results can still be obtained from three or four standard SVFs, each set to highish resonance.

Tony

[kdjupdal]

Here is an idea for a new module:
Why not use the HVM design and convert it into a filterbank with 3 or 4 BP filters, where one can dial in the frequencies of each filter with a pot? This would be a formant filter. The only such filters I've found are in eurorack: the Analogue Systems RS-360, and the Cwejman RES-4. I don't think voltage control is essential but to be able to adjust frequencies is. Formant filters are not only for making the synth say "aah" but it is essential for imitating resonance in acoustic instruments. Something that the HVM module demonstrates beautifully.

Of course, I can just build the HVM but fixed frequencies go against the experimentation that a modular synth lends itself to.

[Synthbuilder]

Why not use the HVM design and convert it into a filterbank with 3 or 4 BP filters, where one can dial in the frequencies of each filter with a pot?

It's not a trivial task to make a Deliyannis filter voltage controlled. Indeed, I'm not sure anyone has done it. That might be an interesting project.

The usual way of making formant style filters, almost certainly like the AS and Cwejman, is using state variable or biquad filters. I'd certainly think about doing such a project at some point. It would, however, be way more complex to build and operate compared to the current HVM. The original half rack version of the HVM did contain, as well as the VP-330 filters, the vocal filter from the Korg Lambda, and a fully sweepable (but not voltage controlled) band pass filter. That was quite a powerful combination.

As an aside I don't believe the AS RS-36 has a +/-12dB response for the band pass outputs as it says in the description. That looks like a bank of three two stage SVFs to me.

Now the real fun with formant filtering is the morphing between the various vowel sounds, ah, ee, ooh, etc. That requires more than just sweeping the filters as the levels of each filter also need to change too. This is where the various software and DSP versions really work well, because to do this in analogue takes a lot of circuitry.