Anyway to get a talkbox/vocoder sound?

Algorithm development and general DSP issues

Moderator: frank

Post Reply
finkster
Posts: 4
Joined: Wed Jun 03, 2009 3:25 pm

Anyway to get a talkbox/vocoder sound?

Post by finkster »

I know this is reaching, but is there anyway to get a talkbox/vocoder type sound by mixing the two inputs one with a mic and the other obviously guitar?
livingston
Posts: 131
Joined: Sun Nov 15, 2009 3:37 pm
Location: New Orleans, LA US

Post by livingston »

Conceptually there's no reason it can't be done, but I think the limit here will be the max program size (128 operations). Here's a block diagram of an analog vocoder:

http://www.paia.com/ProdArticles/vocodwrk.htm

Some of that can be done away with here, but the basic concept of a bank of band pass filters will be necessary, and you won't have enough room in the program to do a vocoder with a lot of bands (more bands=more intelligibility).

So you take your voice input, split it into some bands, and for each band you have an envelope follower. Then you take your guitar input, split it into the same number of bands tuned to the same frequency. Then you use the envelopes from all the voice channels to control the volume of their associated guitar channels.

So
1. Read in ADCL (microphone),
2. bandpass filters
3. For each bandpass output, detect envelope (ABSA, RDFX, WRLX) and zero accumulator
4. Read in ADCR (guitar)
5. band pass filters
6. for each bandpass output, MULX with detected envelopes from before
7. mix the outputs of the volume-modulated guitar bandpasses
8. optionally, you may want to add in the highest frequencies from the voice input, because without this you often lose intelligibility in the voice.
dtech
Posts: 5
Joined: Mon Apr 11, 2011 6:49 pm
Location: Latvia
Contact:

Post by dtech »

I am curious about the fact, that wavefront dsp chip and FV-1 are in some regards somewhat similar (and probably have a comparable performance, too), and there are interesting vocoder programs in AKIRA and other units (airfx) featuring wavefront's chip.

So I wonder if something similar may be written/ported to FV-1 as well?

We are not talking about serious intelligibility stuff like Roland does in VP's, but rather as a toy. Maybe something sounding like a soviet vocoder (sounds and schematics: http://www.ruskeys.net/base/electr26.php), that has 10 rather unselective active filter bands.

The FV-1 design could then be, say, 5-8 wide bands (-12dB/oct) for each bank and it could be possible to do, I think. It will still have a nice voice-like articulation, if bands are properly placed. :)

Digitech did quite a good vocoder processing in about only 230 instructions, as far as I remember, in their old SDISC architecture. So it might be possible to create an Электроника ЭМ-26 grade sound in 120 instructions on an FV-1.

We just need to program it :)

Maybe somebody knows, how it was done in AKIRA rack unit? It has stated in manual to be 40 band!
seancostello
Posts: 74
Joined: Mon Sep 11, 2006 10:04 pm

Post by seancostello »

dtech wrote:I am curious about the fact, that wavefront dsp chip and FV-1 are in some regards somewhat similar (and probably have a comparable performance, too), and there are interesting vocoder programs in AKIRA and other units (airfx) featuring wavefront's chip.
My guess is that the products with vocoders are using the AL3101 chip, as opposed to the AL3201. The AL3101 has the ability to calculate far more instructions per sample than the AL3201 (and the FV-1), but only has a small amount of memory on board. So it is useful for programs with filtering, and can probably handle flanging and choruses, but not pitch shifting or reverbs.

Sean Costello
dtech
Posts: 5
Joined: Mon Apr 11, 2011 6:49 pm
Location: Latvia
Contact:

Post by dtech »

Ah, ok, I finaly managed to find photos of the internals of Akira (it's harder to find anything on my harddrives then to find it in internet). I totaly forgot that Akira has two DSPs - both 3201 and 3102, so no wonder how it can easily do 40 band vocoder and relatively long delay effects at thte same time. There are examples (asm) of a 20band 8channel eq for the 3102 that is not realy far from becoming a two 40band filterbanks (160 non-steep filters are likely cascaded for better selectivity and then divided into two matching banks - and there we get the magical 40 bands).

http://www.dtech.lv/misc/akira_dual_dsp.jpg

While the dsp that does vocoding is significantly different from the spin, we could anyways inspire from the existing design to create one.
Post Reply