Voice FX & cell phone playback

[NotVeryLoud]

I'm working on a voice project for cell phone playback. I need to apply some pretty heavy effects to some of the voices to give them an other-worldly and dreamlike sound. I'm finding this to be a bit of a challenge, as what I create in studio does not at all transfer once it's played back through a voice-mail system. The amount of dynamic compression (or expansion, more likely) is insane. Effects I've mixed in at -15/-20dB below the main voice sound equally as loud, and the noise gates they have in place wreak havoc on any sort of reverb tail, or subtle delay.

I'll workout something eventually, but in the meantime, just wanted to ask for advice on processing voice for phone playback. In particular, voice with reverbs, delays, or other special effects applied to them.

Thanks!

[Rick Sanchez]

Good luck. I would use delay, chorusing or a non-linear/gated reverb.

Anything subtle or reverb with a tail will get destroyed. Just the fact that it's going to be heard on a cell phone pretty much guarantees that it's going to have an otherworldly ghostly and almost unintelligible sound to it.

Forget about dynamics or using any sort of subtle reverb. You can use some rather extreme midrange EQ to get some variety to the voice.

Unfortunately, the telcom and VOIP providers seem to have absolutely no interest in hi fidelity voice quality, even though the technology exists to have superb voice quality over phones. Sadly, 81 year old land line technology still offers the best quality.

[NotVeryLoud]

Thanks Rick,

I may try the EQ thing . . . maybe a sweepable comb filter of some sort.

It's a fun challenge, for sure.

[rafukyo]

try listening to your mix through Speakerphone just to get an idea how it will sound like.

[rfnoise]

Trial and error is probably the best way to go. I think your biggest problems are the way speech compression is being handled these days. Its not just restricted dynamic range or frequency bandwidth. Predictive and time domain coding are kind of like having a "dictionary" of sounds that are used in speech. Encoding interprets original speech into a bunch of descriptions that are sent to the listener's end. Those descriptions are used to assemble something that resembles the original. The coding techniques don't work very well for non-spoken sounds. The "dictionary" doesn't include descriptions for those fancy effects you want to use, so they don't get reconstructed in a predictable way.

Consonant recontruction, where a noise generator and shapers are used to make S, T, SH, etc., can cause some really unusual results. (The original sounds require too much bandwidth to be transmitted well so they are recreated on the receiving end and substituted into place.)

I found this website kind of interesting, especially the examples at the bottom of this page.
Speech Compression

[NotVeryLoud]

Originally Posted by rfnoise

The "dictionary" doesn't include descriptions for those fancy effects you want to use, so they don't get reconstructed in a predictable way.

Wow. Fascinating.
No wonder I'm having so much trouble.

This appears to explain a lot. I've been baffled by the end results when auditioning the final audio over a phone. There are sounds coming out that are so mangled, they bare little resemblance to what was in the original recording.

This also makes auditioning the audio through something like Speakerphone or band limited EQ somewhat useless, as it can't take the encoding, formant prediction, and data compression techniques into account.

I have discovered a technique that seems to get me closer to what I'm trying to achieve . . . with some severe sidechain compression I'm able to duck the reverbs and delays while the main, mostly unprocessed voice is speaking, which keeps it in the clear and intelligible. Early sidechain trigger offsets and slow release times keep the effects from overlapping with the voice. It's not perfect, but it's not too bad.