Hi Ellis,
I feel your point about Denoising and voice intelligibility is important. At
least intelligibility is some sort of objective standard by which we can
measure our success. The problem I see with the standard Denoiser tool
(spectral subtraction) is that it cant lift the partially intelligible voice
out of the noise. It only appears to.
On Dan's example, it seems to lift above
the noise, the louder parts of the voice - but those louder parts were
already intelligible - while leaving the
quieter less intelligible voice parts still buried in the noise. To increase
intelligibility the tool would need to lift the quiet parts of the voice
out of the noise, but it cant. I remember reading an
article from a CEDAR representative explaining this to a Forensics
conference. The tool can only distinguish between soft sounds and loud
sounds, and assumes that louder sounds are wanted and softer sounds
unwanted. At some point we have to decide what to leave in (above the line)
and what to squash (below the line) but in practice
the voice is mixed in with the noise like eggs are scrambled, and there is
no clear line. That explains why in a recording where the voice is barely
intelligible or unintelligible due to broadband background noise, the tool
is useless.
Someone once wrote with some irony that the (spectral subtraction) Denoiser
tool works best when it's least needed...
Regards,
Tim Gillett
Perth,
Western Australia
----- Original Message -----
From: "Ellis Burman" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Saturday, June 23, 2018 11:44 AM
Subject: Re: [ARSCLIST] Processing micro cassette audio using iZotope RX5
I agree with Tom Gillett. The adaptive mode of the RX denoiser takes quite
a bit of time to settle in, and can often over or under process the first
few words of a sentence before it settles. Especially since the noise is
fairly consistent in your case, you should use the manual mode and "learn"
the noise between sentences. This will give a more consistent and
effective result than the adaptive mode.
The de-hum in RX is terrible. The notch filters ring like mad. I love RX,
but that is one feature of it that I find unusable. Luckily, we have Cedar
here, which is crazy expensive, but works like magic, with few artifacts.
If RX is all you have, I'd be very careful using the de-hum. I'd use it as
lightly as possible and listen carefully for the ringing filters. If you
have an FIR EQ with a high enough Q, you might be better off using a few
bands of that, instead of RX de-hum. Be careful removing tones with the
spectral editor too. If you select a long, narrow frequency band and
attenuate it a lot, it will also ring (and pre-ring!) like mad.
You should be able to look at the waveform and determine if the clipping
distortion is playback related. If the clipping is perfectly horizontal,
then it happened on playback, so reducing the volume should help. If the
distortion is not horizontal and looks more embedded into the waveform,
then it likely happened during the recording.
My thoughts on de-noise - The human brain is very adept at hearing through
noise, so I've found that removing the noise doesn't help with
intelligibilty, and often artifacts the audio, making intelligibility even
worse. De-noise is often the most damaging and abused restoration process,
so I'd use it very judiciously, if at all. It can definitely make a track
more listenable, but I have yet to hear it make a track more intelligible.
The human brain is a much more powerful de-noise processor than any plug-in!
Ellis
On Fri, Jun 22, 2018 at 7:42 PM, Tim Gillett <[log in to unmask]>
wrote:
> Hi Dan,
>
> Thanks for uploading the unprocessed audio sample. That really helps. Your
> azimuth alignment seems pretty good. There's a
> bit of azimuth wander but that's to be expected with a slow speed
> microcassette track.
>
> Automatic solutions are not ideal. The best approach is to identify each
> particular type of unwanted noise in
> the recording and then attempt to remove each of them, one at a time, with
> a dedicated tool.
> So with that in mind...
>
> 1. Hum.
> There's a mild hum with a fundamental of around 149 Hz and a total of four
> or maybe five harmonics above that. Probably caused by the cassette
> recorder's motor
> and transport. I was able to greatly remove them with my Acon DeHum tool.
>
> With similar settings you should be able to do the same in RX Dehum.
>
> 2. Random noise.
> Unfortunately the hum is not the main issue. As you mentioned there's a
> lot of
> random noise, and worse of all, right in the critical voice band. That is
> the worst
> kind to remove, often impossible. My approach is always to filter out
> everything below the
> lowest recorded speech frequency, and above the highest. In this case I
> low pass filtered at about 300
> Hz, and high pass filtered at about 3000Hz, both with a Q of about 5.
> It gave a mild improvement in listenability without degrading the wanted
> voice.
>
> Your Corrective EQ tool sounds like the equivalent tool in RX.
>
> 3.Clipping.
> I hear some sort of ugly distortion on voice peaks. Its
> spectrum goes up to at least 10kHz. I doubt it's on the
> microcassette tape. Possibly when you transferred, the playback volume on
> the GE was too high and
> it was clipping its internal amplifier. Try lowering the GE's playback
> volume to where the clipping stops. But dont go too low as there may be a
> low level noise from the GE's amplifier (the hum?) which you want the
> tape's own
> content to mask. Somewhere in the middle is usually the best.
>
>
> You might try RX's Declipping tool but as mentioned I would first check
> the distortion isnt caused by the playback/recording chain.
>
> 4. I didnt attempt to use a Denoiser/Dialogue Denoiser and that was
> intentional.
>
> Hope this helps Dan,
>
> Tim Gillett
> Perth,
> Western Australia
>
>
>
>
>
> ----- Original Message ----- From: "Dan Gediman" <[log in to unmask]>
> To: <[log in to unmask]>
> Sent: Friday, June 22, 2018 11:42 PM
> Subject: Re: [ARSCLIST] Processing micro cassette audio using iZotope RX5
>
>
> Thanks, everyone for all your input so far. Some info that might help
> clarify things. The unit I’m using is a GE micro cassette tape
> player/record, NOT a mini cassette. Second, removing the tape and
> transferring it to a standard cassette is not an option here, as I have to
> return these tapes as is to their owner. Lastly, the unit has a mono
> mini-plug earphone jack and the recording itself is mono. And I’ve already
> adjusted the azimuth to the best of my ability using my ear. I don’t have
> access to a meter to check precisely (unless someone can suggest a
> software
> meter that would work).
>
> As far as RX5, I have used the De-noise module in spectral mode, feeding
> it a sample of the background noise. It has made it substantially better,
> but I’m trying to see if I can make it ever better. I also experimented
> with the presents of the corrective EQ module. I haven’t tried the
> Spectral
> Repair option, because frankly I’m not sure how to use it properly. If you
> think it could help, I’ll go online and find some tutorial videos to bring
> me up to speed. Ditto for any of the other RX5 modules that you think
> might
> help (de-clipping has been suggested, as has de-reverb).
>
> Since I assume most of you have upgraded to RX6, here are the available
> tools in my version of RX5:
>
> De-clip
> De-click
> De-hum
> De-noise
> Spectral Repair
> De-reverb
> Gain
> Corrective EQ
> Channel Ops
> Resample
> Dither
>
> Someone asked if I could upload an audio sample. I’m not sure how to do
> that in a listserv, so I have uploaded a raw, unprocessed sample to Google
> Drive here: https://drive.google.com/open?id=1NuYkyKK6yNXzDRtOA_ZUHuYzf6
> VCAwuK
> <https://drive.google.com/open?id=1NuYkyKK6yNXzDRtOA_ZUHuYzf6VCAwuK>.
> Anything else you think I should try, hearing the file and knowing what I
> have already tried?
>
> Thanks again!
> Dan
>
> Dan Gediman
> 502 299-2565
> [log in to unmask]
> www.dangediman.com <http://www.dangediman.com/>
>
> ---
> This email has been checked for viruses by Avast antivirus software.
> https://www.avast.com/antivirus
>
--
Ellis
[log in to unmask]
818-846-5525
|
