Do Vocoders Have Latency?

Vmusic

Hi,
I have been playing around with some vocoders, for both voice and my violin.

It "feels" like there is some latency in these things (Humanoid). Is that generally the case? I am going by my "feelings", but I well over the 90ms mark.

Maybe I am listening to the wet (vocoder) output in my headphones, and I shouldn't? Will that reduce the latency - just listening to and playing the clean input and turn off the vocoder output?

Thanks
AJ

wim

Is that pitch detector doing pitch detection? If so then that's likely where the latency is coming from. It's not possible to do pitch detection without adding latency.

espiegel123

@wim said:
Is that pitch detector doing pitch detection? If so then that's likely where the latency is coming from. It's not possible to do pitch detection without adding latency.

They may also rely on FFTs which adds latency.

timfromtheborder

Basically any kind of pitch correction involves latency. You need a large enough buffer to detect the pitch in order to transform it. Any given waveform contains a lot of harmonic and inharmonic content so if your buffer is too small, you aren’t getting a representative sample.

wim

I could be getting off the track here, but it seems to me if there's no pitch detection/correction being carried out, a vocoder wouldn't need that type of added latency.

I don't understand how vocoders work very well, but I don't think the actual vocoding part of them uses pitch detection. They split the modulator's audio up into bands, analyze the amplitude of the audio in each band, then apply the same filters to the carrier's audio.

That seems to me like there has to be at least some latency to carry out the analysis, but nothing on the order of what it must take to do pitch detection and correction.

ecou

Humanoid is more than a vocoder. It is also a pitch corrector and a wave table synth.

swarmboy

digital vocoder - yes. analogue vocoder - no.

SevenSystems

There's no reason a vocoder should intrinsically have a lot of latency. The only part in a vocoder that REQUIRES some amount of latency is the envelope followers. So if you can tune them in whatever apps you use, try lowering their attack times or similar parameters.

The other part -- bandpass filters -- should have no more latency than any other bandpass filter in any other system.

Edit: @wim basically beat me to it 😊

SevenSystems

@swarmboy said:
digital vocoder - yes. analogue vocoder - no.

It's not really about an analog vs digital distinction. An analog vocoder that wants to correctly follow the amplitude envelopes in the lower bands will also need more time to do so -- the latency is just introduced in the form of actual electronic components (mostly capacitors) instead of... algorithms 😄

pedro

Even before any complex processing, or look ahead (which introduce latency either by being heavy processing or needing to wait for the look ahead time to even have audio to process), every audio processor does it in chunks (the buffer size) so some latency is unavoidable until they invent some way to “guess” the future audio you’ll feed it

wim

@pedro said:
... until they invent some way to “guess” the future audio you’ll feed it

I've been told that some limiters actually do try to do that in order to try to get by without lookahead. Don't ask me to name which ones, because I don't know. I was just told that by someone way smarter than me that I think I trust.

[edit] fwiw, google AI claims that Logic Pro's adaptive limiter and Izotope Ozone, among others do it. I don't feel like fact checking that.

MonsieurCheval

@wim said:
I've been told that some limiters actually do try to do that in order to try to get by without lookahead.

That’s probably Audio Random Access (ARA). Iirc that was originally made for the studio one + melodyne integration, but I’ve seen some dynamics plugs that use it.