iZotope RX9

[noiseboyuk]

Originally Posted by elangab

For me RX9's DI worked the same, and better, than RX8. It looks like you're cutting a doc or something like that, but something with your test is off for sure. I'm cutting a show and played a bit with 9 on one scene and it was pretty good, when the actors are talking softly I found it to be better than RX8.

Processing time is indeed much slower, hopefully 9.5 will fix it, maybe using more CPU cores or something like that ?

The new Ambience Match is... weird. Didn't work for me AT ALL. It's just random copy/paste with lots of artifacts. Tried to tweak settings, no good.

I'm currently using RX8 for everything, but RX9 DI stand alone plug-in for specific lines.

BTW, there's a setting for RX9's DI module to use legacy algorithm, so you can tick that and it's pretty much RX8, so you don't need to go back and forth if you rather use 9.

Thanks for the report. As a further test I actually recorded the website demo, and it did indeed process exactly as they said - it sounds better, which also bears out your experience. That suggests to me that nothing has glitched in this release version, but that the kind of material that responds well to the new algorithm is quite specific. Presumably they haven't trained the algorithm on a broad enough range of material. My hunch is that the more complex the backing the better the new algorithm handles it (relative to RX8). More even sounds - the easy stuff, you'd think - makes it fall over. As to the speed, I'm not convinced that Best mode really adds anything at all, so Good may be fine for day to day use, and this works at the same speed as 8.

I only tried Ambience Match on one clip so far, but found exactly what you did. It just doesn't work.

If it is the case that legacy DI will work better on some material and RX9 on others, then I'd prefer that checkbox to be right up front rather than buried in settings, as I'll be wanting to switch back and forth all the time. I guess the other option is to have two windows, one 8 and one 9.

But taking stock of all that - all round it still feels like an extremely poor update. If they can retrain the DI algorithm to handle a true variety of material and then update it, then great (though historically iZotope don't really do significant updates in point releases). I do fear though that they will just shrug and say "well it works better on some things than others, and that's why we gave you legacy mode". Add in a non-functioning new Ambience Match mode, it's hardly a compelling update, is it?

[musicman691]

Earlier I mentioned about Product Portal crashing when first opening it? Well that's not quite what's happening. When I first open it it goes into the upper right part of the menu bar and then closes. If I look in the Mac's activity monitor I see it running along with some helper bits & bobs. There's no way to quit it from the menu bar and have to use quit from withing activity monitor.


Anybody else seeing this?

[gives]

Originally Posted by elangab

For me RX9's DI worked the same, and better, than RX8. It looks like you're cutting a doc or something like that, but something with your test is off for sure. I'm cutting a show and played a bit with 9 on one scene and it was pretty good, when the actors are talking softly I found it to be better than RX8.

Processing time is indeed much slower, hopefully 9.5 will fix it, maybe using more CPU cores or something like that ?

The new Ambience Match is... weird. Didn't work for me AT ALL. It's just random copy/paste with lots of artifacts. Tried to tweak settings, no good.

I'm currently using RX8 for everything, but RX9 DI stand alone plug-in for specific lines.

BTW, there's a setting for RX9's DI module to use legacy algorithm, so you can tick that and it's pretty much RX8, so you don't need to go back and forth if you rather use 9.

Good on the DI module to use the legacy algorithm. That will help.

[gives]

Originally Posted by noiseboyuk

Thanks for the report. As a further test I actually recorded the website demo, and it did indeed process exactly as they said - it sounds better, which also bears out your experience. That suggests to me that nothing has glitched in this release version, but that the kind of material that responds well to the new algorithm is quite specific. Presumably they haven't trained the algorithm on a broad enough range of material. My hunch is that the more complex the backing the better the new algorithm handles it (relative to RX8). More even sounds - the easy stuff, you'd think - makes it fall over. As to the speed, I'm not convinced that Best mode really adds anything at all, so Good may be fine for day to day use, and this works at the same speed as 8.

I only tried Ambience Match on one clip so far, but found exactly what you did. It just doesn't work.

If it is the case that legacy DI will work better on some material and RX9 on others, then I'd prefer that checkbox to be right up front rather than buried in settings, as I'll be wanting to switch back and forth all the time. I guess the other option is to have two windows, one 8 and one 9.

But taking stock of all that - all round it still feels like an extremely poor update. If they can retrain the DI algorithm to handle a true variety of material and then update it, then great (though historically iZotope don't really do significant updates in point releases). I do fear though that they will just shrug and say "well it works better on some things than others, and that's why we gave you legacy mode". Add in a non-functioning new Ambience Match mode, it's hardly a compelling update, is it?

I usually just use Ambience Match in Audio Suite. Seem fine for me. I listen to the track and make an adjustment there. I have have better luck for the most part with it in Audio Suite and it's quick after I do a post other fixes on the track so I dont' have to mess with it later. Of course I alway dup each track for safety in case I need to reference things. G

[noiseboyuk]

I've had a couple of responses from iZotope. The gist is that they say Spectral de-noise is better for these sort of cases than Dialogue Isolate. Which is true, it is. But the fact remains that RX8 does it considerably better than 9, which they don't dispute, and it should be able to identify dialogue amid a more constant background sound - most background sounds have a constant aspect to them.

They'd like me to find some other examples of poorly performing Dialogue Isolate and Ambience Match, so let the beta testing begin. If others have problem clips in either module, do forward them on to support so they can gather a number of examples.

Gregory, I'll post some AM examples in the coming days. Worth pointing out that Pro Tools Expert didn't have much luck with the new AM mode either in their review. What I feel it needs to do is essentially provide a seamless version of Dialogue Isolate, but with 100% noise 0% dialogue. Whenever I do that there's always some residual artefacts - were it clever enough to smooth over all those bits and bobs it would be a fantastic tool. As it is, I find the module pretty unhelpful for anything but simple aircon-type chores, I find using Isolate combined with simple cut and pasting frequency selections in Spectral Repair the best way of generating an atmos track with some motion in it.

Speaking of Pro Tools Expert, they concluded that the new de-hum was the star of the show - I haven't actually tried that yet, by and large I find hum relatively easy to fix using other tools so I can't get too excited by that one.

[smurfyou]

Doesn't exactly sound like a must-buy.

Between Acon Dialogue Extract and Accentize Voicegate I'm pretty comfortable with AI denoising. Has anyone compared?

Also if anyone has done a comparison, Absentia is my go-to Hum remover. Better than anything else I try 90% of the time. But it seems like RX9 works well on broadband buzz?

I never went from RX7 to 8 because I didn't notice any improvement from my demo. Thought this one would be worth it...

[noiseboyuk]

I've been through another 5 examples from the same show (again I can't post here, sadly, but I'll forward them on to Support). I didn't cherry pick, just chose 5 random noisy environments usually with mechanical varying noise to see how each fared in Dialogue Isolate. Here's what I found (example 1 is as illustrated in this thread)

WINNERS

1 - jet wash
RX8 (RX9 detected virtually zero dialogue at all)

2 - whine and wash from an electric surfboard
RX8 (RX9 usually better quality, but completely collapsed for one sentence)

3 - whine from a cleaning vehicle
RX9 (RX9 eliminated more background and was better quality)

4 - heavy rattly industrial diesel engine
RX9 (RX9 really nice and clean here, eliminating nearly all the noise)

5 - off-road vehicle
RX8 (Similar to (2) - usually better until it completely collapsed)

6 - all-terrain vehicle
Tie (Very challenging example with a heavy whine ramping up in pitch. 9 generally nicer to listen to, but towards the end lost it and the pitch became overwhelming - at least this could be manually corrected)

While none were quite as awful as example 1, and it clearly won in 2/6 examples, this isn't a good hit rate. A trend seems to be that when it detects dialogue at all, it does sound better quality, and there is less background. However it is overall worse at detecting dialogue in the first place.

Note - all these 5 new examples are similar in terms of battling complex mechanical noise, so it's only one kind of chore for DI. It may fare better at others. Also note - Spectral de-noise works well on one or two of these examples, but the varying pitches means it doesn't work well for the others.

As for Ambience Match, I'm finding exactly what Pro Tool Expert did. I can't get rid of looping artefacts at best. It doesn't work at all if you let it learn a broad selection with dialogue in it, it just chops randomly. And here's the kicker - if you have enough of a clean selection of atmos to train it well... then you'd just use the clean bit of atmos! It fundamentally doesn't do what you need it to, to give it a single mic with atmos and dialogue, and produce a useable and consistent atmos-only output.

(if anyone does want to hear the other examples, drop me a PM).

[gives]

Originally Posted by noiseboyuk

I've had a couple of responses from iZotope. The gist is that they say Spectral de-noise is better for these sort of cases than Dialogue Isolate. Which is true, it is. But the fact remains that RX8 does it considerably better than 9, which they don't dispute, and it should be able to identify dialogue amid a more constant background sound - most background sounds have a constant aspect to them.

They'd like me to find some other examples of poorly performing Dialogue Isolate and Ambience Match, so let the beta testing begin. If others have problem clips in either module, do forward them on to support so they can gather a number of examples.

Gregory, I'll post some AM examples in the coming days. Worth pointing out that Pro Tools Expert didn't have much luck with the new AM mode either in their review. What I feel it needs to do is essentially provide a seamless version of Dialogue Isolate, but with 100% noise 0% dialogue. Whenever I do that there's always some residual artefacts - were it clever enough to smooth over all those bits and bobs it would be a fantastic tool. As it is, I find the module pretty unhelpful for anything but simple aircon-type chores, I find using Isolate combined with simple cut and pasting frequency selections in Spectral Repair the best way of generating an atmos track with some motion in it.

Speaking of Pro Tools Expert, they concluded that the new de-hum was the star of the show - I haven't actually tried that yet, by and large I find hum relatively easy to fix using other tools so I can't get too excited by that one.

Thanks Guy. I never heard back from anybody there yet, but a lot of tweaking works, but it's trial and error sometimes. I have had pretty good success using DeRustle in some cases and sometimes don't even use Dialogue Isolate and if I do, I use most of the functions sparingly.
I save a lot of presets that are working for each project that are specific to it. What I hate is some of the clients. BIG ones that evaluate their project in headphones, (God knows which ones,Beats etc.), and then comment in 10 different directions. Had that last month on a web series last month or so. It turned out great, but the producer finally had to shut down the back and forth because they were contradicting each evaluation on Frame IO over and over again. PLUS, some of the dialogue that had "echo" in it was because they had NO coverage with separate Lav and Mic. They had them summed together and it was a real mess. I got more or les sounding good. There were other things, but anyway, you would think that these companies that make $$, would do a better job with sound on location?-:). then I had some great projects right or during that which seemed like Childs play after that encounter. I'll leave you with that on a Sunday as I head up to do a favor for friend. P.S. Spectral Repair is my friend as well.

G

[gives]

Originally Posted by noiseboyuk

I've been through another 5 examples from the same show (again I can't post here, sadly, but I'll forward them on to Support). I didn't cherry pick, just chose 5 random noisy environments usually with mechanical varying noise to see how each fared in Dialogue Isolate. Here's what I found (example 1 is as illustrated in this thread)

WINNERS

1 - jet wash
RX8 (RX9 detected virtually zero dialogue at all)

2 - whine and wash from an electric surfboard
RX8 (RX9 usually better quality, but completely collapsed for one sentence)

3 - whine from a cleaning vehicle
RX9 (RX9 eliminated more background and was better quality)

4 - heavy rattly industrial diesel engine
RX9 (RX9 really nice and clean here, eliminating nearly all the noise)

5 - off-road vehicle
RX8 (Similar to (2) - usually better until it completely collapsed)

6 - all-terrain vehicle
Tie (Very challenging example with a heavy whine ramping up in pitch. 9 generally nicer to listen to, but towards the end lost it and the pitch became overwhelming - at least this could be manually corrected)

While none were quite as awful as example 1, and it clearly won in 2/6 examples, this isn't a good hit rate. A trend seems to be that when it detects dialogue at all, it does sound better quality, and there is less background. However it is overall worse at detecting dialogue in the first place.

Note - all these 5 new examples are similar in terms of battling complex mechanical noise, so it's only one kind of chore for DI. It may fare better at others. Also note - Spectral de-noise works well on one or two of these examples, but the varying pitches means it doesn't work well for the others.

As for Ambience Match, I'm finding exactly what Pro Tool Expert did. I can't get rid of looping artefacts at best. It doesn't work at all if you let it learn a broad selection with dialogue in it, it just chops randomly. And here's the kicker - if you have enough of a clean selection of atmos to train it well... then you'd just use the clean bit of atmos! It fundamentally doesn't do what you need it to, to give it a single mic with atmos and dialogue, and produce a useable and consistent atmos-only output.

(if anyone does want to hear the other examples, drop me a PM).

Hey Guy, Did you try using it in AudioSuite and select the function there? I am curious if it's just in exporting to the full suite issue. Typically I only use it in Audio Suite and build it there in Protools on another track and use as needed for the most part.

[gives]

Originally Posted by elangab

For me RX9's DI worked the same, and better, than RX8. It looks like you're cutting a doc or something like that, but something with your test is off for sure. I'm cutting a show and played a bit with 9 on one scene and it was pretty good, when the actors are talking softly I found it to be better than RX8.

Processing time is indeed much slower, hopefully 9.5 will fix it, maybe using more CPU cores or something like that ?

The new Ambience Match is... weird. Didn't work for me AT ALL. It's just random copy/paste with lots of artifacts. Tried to tweak settings, no good.

I'm currently using RX8 for everything, but RX9 DI stand alone plug-in for specific lines.

BTW, there's a setting for RX9's DI module to use legacy algorithm, so you can tick that and it's pretty much RX8, so you don't need to go back and forth if you rather use 9.

Thanks for the tip on legacy algorithm-:)

G