Topic: Processing amplified speech

[WiFiJeff]

I am working on tapes from some friends of a piano recital with poetry readings interspersed.  The hall was fairly reverberant, but the piano parts are "normal" to edit; the poetry however was amplified, and both some direct voice and lots of speaker voice plus reverb are on the tape.  I have tried some iZotope Rx tools, both "dereverb" and "deconstruct" and find that I get lots of artifacts with no great improvement in intelligibility.  Anyone know of any useful tools that might help here, or pointers on settings for Rx that might improve my results?

Jeff

[page]

Are they speaking while there is playing or just speaking?

Second, are they centered in the stereo image or pushed slightly left/right?

[stevetoney]

Sometimes when I'm tracking a show and trying to figure out the setlist, a song title will be announced between song.  If I can't figure out what is said, I'll EQ varying degrees of high end in, which seems to crispen the speech and sometimes helps the intelligibility.

[Gutbucket]

I find the time delay between the direct vocal arrival and the reinforcement from the PA a difficult thing to compensate for.

If it wasn't for that aspect I'd suggest simply EQing it with a healthy presence range boost and possibly a bit of an upper-bass/lower-mid scoop if it's vocal recital only without music behind it.  If accompanied, I'd probably try a similar strategy with a milder curve to balance the benifit to the vocal against any detrimental effect on the music.  That can help speech clarity and intelligibility directly from a tonal perspective, and partly compensate for excess room 'verb even though it does nothing to actually reduce the level of it.  But that no news to you guys I’m sure.

EQ doesn't do anything for the difficult 'slapback echo' problem though.  I frequently encounter that with classical material in a particular hall with spoken segments before the piece by the composer, conductor, or performer who are typically standing 10' away from where I sit, while the PA speakers are behind them something like 50' away, soffit mounted high in the wall behind the stage.  Fortunately it only affects the spoken segments since the performances are usually fully acoustic or occasionally for things like Steve Reich pieces, visiting amplified ensembles or electronic organ material they'll place mobile PA speakers on-stage about equidistant with the acoustic sources.  Farther back in the audience the direct sound of their speach is probably attenuated enough so that the PA dominates and the problem is not apparent for most patrons.

A ‘de-slap / de-verb’ algorithm would be something of a golden for live recording egg if it worked transparently enough.  The tools in Rx you mention come to mind and have intrigued me, but I’ve never tried them and suspected that they just weren’t clean enough to be of much use for music (if perhaps quite useful for something like forensic audio).

If any of you guys find one that does works decently for music, please let me know.

[page]

The tools in Rx you mention come to mind and have intrigued me, but I’ve never tried them and suspected that they just weren’t clean enough to be of much use for music (if perhaps quite useful for something like forensic audio).

RX is a good tool and it costs accordingly. I had a friend who wasn't using his license for a year and I borrowed his. I sort of miss it in some ways as you can accomplish some down right amazing things.

Anyway, as others have said, you want to EQ it on the top end for intelligibility reasons (8-10k works well to emphasize). Do a cut around 400hz, and anything under about 150hz which will help with both intelligibility and help knock out some of that washy verb. There is another segment you can cut but I'd have to be sitting at my desk at home to see the notes. Second, if the vocal is centered in the image, you can use a mid-side plugin to emphasize the center (as the reverb should be different on each side at a mathematical level). Don't overdo it though, I think that's a dangerous (but powerful) tool when using it on anything other than true M/S recorded work. That's what I start with at any rate. Anything else would require a sample.

[Gutbucket]

Good call on a moderate M/S ratio tweak, I didn't consider that angle.  It might help if the reverb is predominantly decorelated between channels, probably not so much if the reverb is moslty shared monophonic information.  The mic technique used is likely to make a significant difference with that.

EQ is easy enough to play around with to find frequency regions best targeted as long as you're patient.  When I mentioned presence region I was thinking a bit further down like ~1-3kHz for the heart of the clarity and intelligibility stuff.  I think of 8-10kHz as being more of a vocal silibance and brightness region, with space and air above that.

While adjusting that I'd focus primarily on its effect on the vocal itself, listening to balance whatever was achieved there against whatever it harmed in other areas like the room sound, other noises (presuming no music). When doing any low cutting or low-mid conturing I'd likely switch mental focus to listening for the effect that has on reducing the influence of the reverberation instead, balancing those gains against any harm it does to the vocal quality. 

[page]

The 8-10khz region helps with hard consonants, I got the idea from a number of "vocal" mics which all have the spike somewhere over 7k but under 11. Played around wtih it a little bit and found depending on the voice/pa/room/etc that you can do some stuff there that will help with intelligibility.

For normal EQ, I totally agree, listen to the room. I'd take it one step further and say "listen to what's important." 4/5 times, the room reverb and make/break a recording. I'm not sure how pertinent speech would be though, one could mix EQ fades** and handle it that way, but it just depends on how valuable that speech is. I once did a pianist and monologue where I had to do a lot of work on the vocal, but none on the piano. In that environment, the jarring change of EQ was preferable to losing the historical performance.

** Render the EQ'ed edition of the edits, and take that and do fades surrounding where you want to keep the EQ. Now take the original one and do the inverse (fade out and in) leaving holes where you want the EQ to stay. Now mix the two. that creates an ambiance transition instead of a hard edit.

[WiFiJeff]

Are they speaking while there is playing or just speaking?

Speaking is between the piano pieces, thank goodness for that.

Second, are they centered in the stereo image or pushed slightly left/right?

Pretty much centered, there were speakers on each side of the hall.

Thanks.

Jeff

[WiFiJeff]

Thanks all for the suggestions.  I ended up using the odd M/S step to start, in Wavelab with the VST plug-in Stereo Tools M/S module, then doing some EQ.  I then took those files into RX3 Advanced and "dereverbed" them, letting the Dereverb train itself on a section (playing around with the parameters didn't lead to much).  Then I notched out very low frequencies and used RX3 to get rid of a lot of audience noise in detail.  It ended up being much easier to listen to, but not much more intelligible.  I will pass the processed files on to the original taper, and see if he is happy or outraged at the result.

Jeff

[Gutbucket]

Hopefully pleased if not outrageously happy.

Currious how those "deverb" tools work under the hood.  I'll play around with them someday.. your experience bolsters my suspicion that they probably aren't yet a golden path to audio salvation for overly roomy AUDs at this point.