Another attempt to fix this awful file

[Kathy Smith]

Originally Posted by Greg Miller

Sorry for the confusion about the hum (or non-hum). I was just looking for an explanation for the strange vocal distortion that I seem to hear in several spots.

How sad that there is no earlier generation or better version of this file. :-(

Out of curiosity, was it recorded with a wired mic, or with wireless? I hear a little noise burst, just after "... all the way around it" and I wondered whether that was an RF hit. Or maybe it's a clothing rub ... the white noise really obscures what's going on there.

Wireless mic. I agree it's hard to hear anything under the white noise.

[Greg Miller]

Aha, I thought so.

I wonder whether that white noise wasn't the result of a problem with the wireless system. I know you said earlier it related to phantom power, and I don't say you're wrong. But a weak RF signal, or a receiver with NO signal to lock onto, can also produce white noise. So, as usual, I'm curious.

At any rate, it's a shame we can't go back and undo the problem. Shoot the recordist, (s)he should have been monitoring on headphones!

[Bernie Beaudry]

Originally Posted by Kathy Smith

I agree Greg's file sounds best.

Hi Kathy,
Greg's sample in post #7 does sound pretty good but its only the first :05 seconds. Listen to my post number 10. I did the whole file and listening to both on headphones to me they both sound similar. It's a little lower in volume than it was but that's easily fixed. I tried to get the voice to sound more natural without the voice modulating the remaining background noise.
I'd like to hear your impression of the result so I can see if my approach was useful.
Best
Bernie

[Christopher Young]

Originally Posted by Kathy Smith

I agree Greg's file sounds best.

+ 1 on that. Greg's is the best.

[Bernie Beaudry]

To everyone on this thread. Please take a listen to my example in post # 10. I'm learning RX and I've been an audio person for 38 years. At least 30 of that as a location sound recordist. To my ear, with headphones on, my effort is very similar to Greg's. I'd like some constructive criticism so I can determine how effective my approach was. My pervious try was done with my computer desk speakers and I didn't hear how many artifacts were left.
I know what my ears are telling me, I'd value your perception.
Bernie

[Greg Miller]

Since you've asked everyone for advice and comments, here are mine.

Your file sounds a lot noisier than mine. The RMS level of your noise is about -24dB lower than the overall RMS level of the entire file. With my sample, the same S/N ratio is about -35dB.

I think my dialog is somewhat clearer, too, but I can't put a number on that. (I'm listening on Senny HD-280 Pros. Different cans will sound different.)

Admittedly, my example was done more or less manually, a syllable or two at a time, with a lot of trial and error to find the best compromises. It was not a matter of setting a few parameters and clicking the "process" button for the entire file. It involved listening closely to each small piece of audio, thinking about what I heard, figuring out the best way to fix that section, and then using the software to achieve that goal. I spent the better part of an hour on those few seconds of audio.

Is that a reasonable approach for a file like this? That depends on the value of the audio. If it's a world leader making an exceptional statement, yes. If it's your family's home videos, no. If it's a lady talking about flowers ... I can't answer that. I don't know whether this is sync sound, or a voiceover. Maybe a re-shoot or ADR would be a better solution. Maybe a "quickie" with worse results would be an adequate compromise.

I was simply trying to demonstrate that the file can be made to sound decent, if the right things are done.

[Bernie Beaudry]

Originally Posted by Greg Miller

Since you've asked everyone for advice and comments, here are mine.

Your file sounds a lot noisier than mine. The RMS level of your noise is about -24dB lower than the overall RMS level of the entire file. With my sample, the same S/N ratio is about -35dB.

I think my dialog is somewhat clearer, too, but I can't put a number on that. (I'm listening on Senny HD-280 Pros. Different cans will sound different.)

Admittedly, my example was done more or less manually, a syllable or two at a time, with a lot of trial and error to find the best compromises. It was not a matter of setting a few parameters and clicking the "process" button for the entire file. It involved listening closely to each small piece of audio, thinking about what I heard, figuring out the best way to fix that section, and then using the software to achieve that goal. I spent the better part of an hour on those few seconds of audio.

Is that a reasonable approach for a file like this? That depends on the value of the audio. If it's a world leader making an exceptional statement, yes. If it's your family's home videos, no. If it's a lady talking about flowers ... I can't answer that. I don't know whether this is sync sound, or a voiceover. Maybe a re-shoot or ADR would be a better solution. Maybe a "quickie" with worse results would be an adequate compromise.

I was simply trying to demonstrate that the file can be made to sound decent, if the right things are done.

Thanks for taking the time Greg! I didn't do an analysis so I appreciate what you're saying. My goal was to have the voice sound as natural as possible, and not have the background pump up and down. The tradeoff was leaving more noise in the background. I used eq, deconstruct, noise reduction, and declip on selected areas. I didn't choose some parameters and hit process as you implied. I'm listening on Sony 7506s. It took about an hour to do the whole file, but of course I was building on the previous try I did. I can hear that my example also has a little less presence and volume. I'm going to give it one more shot and see what else I can learn.
I appreciate the feedback.
Bernie

[Greg Miller]

Bernie,

Unfortunately, I do hear noise pumping up and down in your file. (e.g. from 0:20.0 to 0:24.5) Admittedly, there are slight noise changes in my file, too, albeit at a lower level.

The 7506 has a well-known midrange boost. Some people like them for monitoring on a live shoot, because they tend to accentuate any noise that might be picked up by the mics ... so you can fix the problem instead of recording a noisy track. I definitely do not trust them for EQ purposes (or for NR evaluation in this case), because they give a false impression about intelligibility. I especially would not mix on them, because they tend to make the voice stand out over FX and music; play the same track on good speakers (or on Senny 280s) and you might discover that the voice gets lost in the mix.

Today's audio software is miraculous, but you still need to tell it what to do. With an average file you can often "set it and forget it." But with something this bad, you really need to carefully set the parameters, and then tweak them a lot. I personally could not find any settings that would let me get "really wonderful" results for the whole file; I needed to tweak it, almost syllable by syllable. Maybe you will have better luck.

And, again, what is "good enough"? Only Kathy can answer that question about this file.

Greg

[Bernie Beaudry]

Originally Posted by Greg Miller

Bernie,

Unfortunately, I do hear noise pumping up and down in your file. (e.g. from 0:20.0 to 0:24.5) Admittedly, there are slight noise changes in my file, too, albeit at a lower level.

The 7506 has a well-known midrange boost. Some people like them for monitoring on a live shoot, because they tend to accentuate any noise that might be picked up by the mics ... so you can fix the problem instead of recording a noisy track. I definitely do not trust them for EQ purposes (or for NR evaluation in this case), because they give a false impression about intelligibility. I especially would not mix on them, because they tend to make the voice stand out over FX and music; play the same track on good speakers (or on Senny 280s) and you might discover that the voice gets lost in the mix.

Today's audio software is miraculous, but you still need to tell it what to do. With an average file you can often "set it and forget it." But with something this bad, you really need to carefully set the parameters, and then tweak them a lot. I personally could not find any settings that would let me get "really wonderful" results for the whole file; I needed to tweak it, almost syllable by syllable. Maybe you will have better luck.

And, again, what is "good enough"? Only Kathy can answer that question about this file.

Greg

All good feedback Greg, thanks! You're right about the Sony's. I'll give the 280s a try. I should also get my good monitors working too. I'm just at the beginning of this software journey so I appreciate your input and advice. This was a very good learning file to play with, so thanks to Kathy for asking for help on it. By the way are you also on RX, or are you using something else or a combination of RX and something else? By the way that whoosh sound in that other file was also a very, very tough one. I was playing with that one too, and its really hard to even see the sound. You did a good job on it. Have you ever used Zynaptiq Unchirp on anything? I'm wondering it it would work on some of the artifacts left behind from noise reduction.
BB

[Greg Miller]

Bernie,

I think you'll like the 280 Pro. Much flatter than the 7506, and better isolation, too. I even find them to be more comfortable.

Everything I've done on this file was done with an old version of Audition, or with Cool Edit Pro (they are largely the same program). I have switched back and forth, depending on which machine was nearby; I don't really remember which program I used for which example.

Greg

[Kathy Smith]

I'm using Sennheiser HD 280 and hear a huge difference between Greg's file and Bernie's. I also hear sibilance in Bernie's file, which I don't hear in Greg's.

[Kathy Smith]

Here is my attempt. What's your opinion of this file?
https://www.dropbox.com/s/a1q4h4kzz2...d_Fix.wav?dl=0

[Greg Miller]

It's an interesting compromise (anything with this source file will be a compromise). Certainly not bad.

I think you've applied a fair amount of EQ, because the bottom end sounds fuller than the original.

I still hear some artifacts within/under the voice, although they are not "burbly" as artifacts often are. They are less objectionable to me than artifacts on some other peoples' attempts.

I seem to hear the frequency response changing, depending (seemingly) on the loudness of the source material. Listen from 0:29.4 to 0:36.6. The first few words ("I wanted to ...") sound nice and open, but then the HFs seem to gradually disappear and the sound becomes muffled. Then the next phrase ("discover I guess ...") opens up suddenly, and then gradually becomes muffled again. So whatever process you're using for NR seems to be constantly changing the bandwidth, or at least the response curve, depending on the overall level of the audio.

(You can also clearly hear the artifacts within this same 7-second section.)

And yours sounds fuller than mine, because I used bandpass filtering to get rid of some of the noise at the top and bottom end of the frequency range ... frequencies that aren't necessary for intelligibility.

What do you think? Is it usable for your purposes? (I have no idea what your final product will be.)

[Bernie Beaudry]

Originally Posted by Kathy Smith

Here is my attempt. What's your opinion of this file?
https://www.dropbox.com/s/a1q4h4kzz2...d_Fix.wav?dl=0

Here's one more attempt from me. I ordered some HD 280s but in the interim I thought I'd see what else I could do. Let me know what you think.
Bernie
https://www.dropbox.com/s/j0t4dgawhh...test1.wav?dl=0

[Kathy Smith]

Originally Posted by Greg Miller

It's an interesting compromise (anything with this source file will be a compromise). Certainly not bad.

I think you've applied a fair amount of EQ, because the bottom end sounds fuller than the original.

I still hear some artifacts within/under the voice, although they are not "burbly" as artifacts often are. They are less objectionable to me than artifacts on some other peoples' attempts.

I seem to hear the frequency response changing, depending (seemingly) on the loudness of the source material. Listen from 0:29.4 to 0:36.6. The first few words ("I wanted to ...") sound nice and open, but then the HFs seem to gradually disappear and the sound becomes muffled. Then the next phrase ("discover I guess ...") opens up suddenly, and then gradually becomes muffled again. So whatever process you're using for NR seems to be constantly changing the bandwidth, or at least the response curve, depending on the overall level of the audio.

(You can also clearly hear the artifacts within this same 7-second section.)

And yours sounds fuller than mine, because I used bandpass filtering to get rid of some of the noise at the top and bottom end of the frequency range ... frequencies that aren't necessary for intelligibility.

What do you think? Is it usable for your purposes? (I have no idea what your final product will be.)

Thanks Greg. It will have to do for my purpose. I do hear the sound opening and and then becoming muffled and then opening again etc. I need to go back and hear whether it's the NR doing that or the speaker herself.