Processing steps for telephony voice prompts...

[minister]

hi tobias,

do you have the conversion software? do you have a way to put the sounds onto a phone system? like, your voicemail or something?

my suggestions are to convert the sounds and listen to them. i even go so far as to transfer them to voice mail (which is not always identical to what your application will be but it gets you in the football stadium). i have done several cel-phone audio tours for art museums around the united states and i knwo the producer very well. we always post test files.

point is, you need to process the sounds after you know the end product (kinda like making mix adjustments while monitoring theough the Dolby Matrix). you may find that you need to make things a little brighter than you normally would. and maybe a little 'maximized' than you normally would. but be careful of sibilance. the phone system does not represent much under 200Hz nor much over 4kHz. women come off sounding more natural, yet, can easily become shrill and men can sound like baritones turned tenor. each voice is different.

i like to EQ and limit a bit more than i normally would and i often throw on a MAXIM or BBE-type or INFLATOR plug-in on there to. be careful not to go crazy with these, they can get harsh fast and your ears adjust to it... come in the next day and go, "what the hell was i doing???"...

BarbaBatch form Audioease is probably the best convertor batch convertor software for the MAC. but Sound grinder is really good too and much cheaper!

[to-pse]

Originally Posted by minister

hi tobias,

do you have the conversion software? do you have a way to put the sounds onto a phone system? like, your voicemail or something?

I use CoolEdit to dither & convert from 44.1khz/24 to 8khz/16bit.
I already did this in the past with prompts I got from recording-
studios. Works like a charm & the CoolEdit-SRC has a pretty good
quality - so no problem & questions there...

i like to EQ and limit a bit more than i normally would and i often throw on a MAXIM or BBE-type or INFLATOR plug-in on there to. be careful not to go crazy with these, they can get harsh fast and your ears adjust to it... come in the next day and go, "what the hell was i doing???"...

I'd like to get into the recording & post-processing (eq, dynamics...) of
such prompts and so far I really don't know what processing is done
"by the pros" for such an application (telephony-systems with limited
bandwidth).

I have done some music vocal recordings before & some mixes together
with instrumental tracks. But just voice alone is something completely
different, because the vocals just need to "sound on their own" without
any background... so any hints on what the post-processing steps for
such recordings normally include would be highly welcome...

Tobias

[lazzaro]

Probably not a concern since you're in Germany,
but if the voice prompts may be played back
on CDMA networks, you should try to preview it
on a CDMA phone. The Qualcomm PureVoice
codecs do very different things to certain voices,
as compared to GSM codecs.

[minister]

your tools are fine.

as i mentioned in my post, you need to listen to the files on something that resembles a phone. a bad speaker, a phone....

here is a link for Plain Old Telephone Audio (POTS)

http://www.rane.com/note150.html

make the voice sound good and rich and full and controlled without being TOO compressed. don't boost 125Hz too much becassue it will be gone anyway. and don't make it too bright either. a good natural tone is what you want. make sure the compressor doesn't pump! but know what the end result will be : bandwidth limited.

[Mundox]

For Telephony work (IVRs) I generally roll off above 10k and below 120Hz w/ a pretty sharp curve, sometimes boost around 2k, and compress/limit the hell out of it.

[charles maynes]

Originally Posted by Mundox

For Telephony work (IVRs) I generally roll off above 10k and below 120Hz w/ a pretty sharp curve, sometimes boost around 2k, and compress/limit the hell out of it.

With the EQ- A peaking filter can be a good way to go- the EQ the phone company uses has nothing to do with "pleasing" it has more to do with, if you will pardon the vernacular- "Bitchslap"

To make it even more convincing, you can do some nulling eqs which will exaggerate the distortion- IE do a very tight very resonant filter at about 6 to 8khz with an anti-filter following it.

though just reamping through a TV speaker works really well too- I actually did that for some stuff on "Catch Me If You Can" where I needed to process some ADR to fit a Kinescope'd tv program from the 50's.

charles maynes

[Mundox]

Hey Charles,
I think you mis-understood the original question Tobias asked.
He's after advice for how to process files to go on Phone systems.
Your advices are for making them sound like it's coming down a phone line.
Though very useful in their own right.

I actually use "cosmonaut voice" plug for that purpose. Works well.

[charles maynes]

Originally Posted by Mundox

Hey Charles,
I think you mis-understood the original question Tobias asked.
He's after advice for how to process files to go on Phone systems.
Your advices are for making them sound like it's coming down a phone line.
Though very useful in their own right.

I actually use "cosmonaut voice" plug for that purpose. Works well.

I guess rest is a valuble thing... and here I thought it was cool that the conversation had included codecs-

sorry

charles maynes

[JimmySX]

Recordquality as high as possible. After that compress the hell out of it and use a filter to get rid of the lows. Boost around 4K. Then run the whole bunch thru a L2 or something similar and voila you're done!!!

Convert with CoolEdit (which is really a cool app for telephone stuff) and you're done

[to-pse]

Hi again,

first of all thanks to everyone for their suggestions... I will be trying out
some different stuff in the next couple of days...

Concerning "emulating" a phone-call I have seriously considered writing
a small VST which bandwidth-limits a signal to 100-3100 Hz (either by
filtering or downsampling) and does an actual codec-conversion from
linear to G.711 (u-Law or ALaw) and back to linear to get "the real thing".

Don't know if this is worth the hassle, or if just applying EQs is a good
enough approximation of the telephone-line...

Originally Posted by JimmySX

Recordquality as high as possible. After that compress the hell out of it and use a filter to get rid of the lows. Boost around 4K. Then run the whole bunch thru a L2 or something similar and voila you're done!!!

What do you exactly mean by "compressing the hell out of it"? High
compression-ratio? I would really appreciate a short hint what kind of
attack/release and compression-ratios could be useful... just to get a
starting-point from which to go on and experiment a little...

What is the reason for having the L2 as the last step, if I have already used
a large amount of compression?

Convert with CoolEdit (which is really a cool app for telephone stuff) and you're done

I have been using CoolEdit for years now - it is very intuitive to use and the
sample-rate-conversion is really good... I'm currently tracking the voices using
Cubase SX 2 with 44.1/24, applying VSTs there, rendering it out to 44.1/16 and
then doing the final cuts and SRC in CoolEdit...

Tobias

[edboy7]

Originally Posted by to-pse

Hi everyone,

I'm currently doing some recordings that are supposed to
be used as voice-prompts for a telephony-system (8khz
samplerate).

Recordings are either done in 44.1 or 48khz sample-rate,
with either an AT3035 or an MXL V69M mike going straight
into a Sydec Mixtreme Preamp/Converter-Box.

So I'm looking for advise on what to do for post-processing
(apart from converting to 8khz & dithering in the last step).

So if anyone can give me few hints on how to post-process
those recorded files, I'd really appreciate that...

Best regards,
Tobias

We're using Wavelab for post proccessing and got no client complaints whatsoever.

[jowillie]

OK gang. I just recorded an answering message for a big Winery. I used some of your suggestions. I'll let you know what the client thinks.
Thanks
WE

[Mundox]

Originally Posted by to-pse

What is the reason for having the L2 as the last step, if I have already used
a large amount of compression?

You have to get the peaks flattened out and normalize the files, otherwise there will be great volume differences between messages. Get as much dynamics out of it as possible. We learned it the hard way believe me, people complain about files not being loud enough unless it's to the point of ANNOYINGLY LOUD!
Also strap in a De-esser at 5k shelf.

[JimmySX]

Originally Posted by Mundox

You have to get the peaks flattened out and normalize the files, otherwise there will be great volume differences between messages. Get as much dynamics out of it as possible. We learned it the hard way believe me, people complain about files not being loud enough unless it's to the point of ANNOYINGLY LOUD!
Also strap in a De-esser at 5k shelf.

Thx Mundox for pointing that out