Log in
Log in

or
Create an account

or

Thread Trying to clarify voices

  • 9 replies
  • 2 participants
  • 1,066 views
  • 2 followers
1 Trying to clarify voices
Hey all!

Hoping for a little helping hand... I have an audio recording taken with a digital voice recorder from inside a bar that occasionally got pretty loud in the background (background voices and music). I don't expect perfection, I'm only looking for coherence. The loudest voice on the recording, the female I was speaking with, is the voice I need to hear. I'm trying to recall the entire conversation, but I already will recall what I said if I can just hear what she said. Does anyone have any good tactics for dealing with this? I've had mixed results with everything I've tried on things I've read online, but I'm still having so much trouble understanding the voices. Again, not looking for perfection here, just coherence. I'll attach a sample of one of the better portions of the audio that still contains background chatter, but clearly gives the voice I need to try and isolate.

If anyone can help, or even point me to someone who CAN help, that would be great! Thank you all so much!

Btw, I have audacity and Izotope RX5 to assist in this, I am just NO sound engineer. Thanks again!https://drive.google.com/open?id=0B5vTDz4l9lsWNmJFWEVrdzF0RGs
2
I'm unable to open your audio file. Can you convert it to a WAV or MP3 and repost? Thanks!
3
Absolutely. It's originally in .wav. I'll make it now and repost. Thanks!
4
OK, here is a 5 minute sample. Its towards the beginning, when the bar was still pretty quiet, everything after this gets far worse. But if I can find a way to process this kind of section, maybe I can improve the later parts. Thanks again!

https://drive.google.com/open?id=0B5vTDz4l9lsWbklSek0zR0NyMjg
5
I listened to the file, and messed around with it in RX5. The problem that you have is that the background noise is louder than the source you're trying to hear, much of the time. That's referred to in the audio-restoration world as being "underwater." I was able to bring out her voice somewhat by first using the De-Noise module, in manual mode, and selecting a section for the program to "Learn" where her voice was not present. Then I used the EQ, high-passing much of the low frequency spectrum, and boosting in the upper meds (around 2K) which helped bring out her voice somewhat. That said, I couldn't get her voice to be understandable throughout. The recording is also heavily clipped, which doesn't help. Since you have RX, maybe try experimenting more with what I was trying. You might also consider writing to iZotope to see if they can give you some tips for using RX to solve this. It's a real tough one, though. Good luck. :bravo:

[ Post last edited on 05/18/2016 at 08:10:59 ]

6
Much appreciated! Just a follow up- do you use a specific level of noise reduction? I was initially trying to just kill the background with maximum reduction, however I've learned now that it seems to work best lowering it in smaller increments of around -8dB. I have no handle on what the different settings do... so I've just been trying random things to see if I can mess with it.

Also, what does it mean that the voices are heavily clipped? Is there something I can do about that?

I had also mentioned, I have two clean portions with basically no background noise, which are the drive to the bar and the drive home... is there any way for me to utilize those clean recordings to pull forward the voices in the rest of it? Or, alternatively, find a way to eliminate their voices from the rest of the recording, and then somehow delete everything that is left from the full recording? Basically treating their voices as the noise, then using the recording of only the background noise and somehow use that to delete it all from the original recording?
7
Quote:
Much appreciated! Just a follow up- do you use a specific level of noise reduction? I was initially trying to just kill the background with maximum reduction,

The downside of too much broadband noise reduction is that it adds artifacts that make a voice sound like it's underwater (aka "musical noise"), but considering that your goal is only to be able to understand what she's saying, then there's no reason not to use as much of the reduction as possible. I suggest finding a small section where her voice is not present (but the background sounds are), and using the De-Noiser's "Learn" function to capture that as a noise profile. Then apply as much reduction as you can without affecting intelligibility. You can also mess around with the Threshold parameter, but usually the one that gets set via the Learn function is best.

Quote:
Also, what does it mean that the voices are heavily clipped?

Clipped means that it recorded at over 0 dB, and sounds distorted. You can tell a clipped waveform because the top of it is flat, because the very top was "clipped" off when it hit the maximum. There isn't much you can do. RX does have a De-Clip function, but I haven't found it to be consistently effective. Still, it's worth trying. iZotope has some very good tutorial info on its website. Here's a link to the RX Audio Cookbook, which is a new tutorial section dedicated to RX: https://rxcookbook.izotope.com/?_ga=1.216444246.1342937587.1398213946

Quote:
I had also mentioned, I have two clean portions with basically no background noise, which are the drive to the bar and the drive home... is there any way for me to utilize those clean recordings to pull forward the voices in the rest of it? Or, alternatively, find a way to eliminate their voices from the rest of the recording, and then somehow delete everything that is left from the full recording? Basically treating their voices as the noise, then using the recording of only the background noise and somehow use that to delete it all from the original recording?

The only methods I know of for getting rid of the background noise are what I've already suggested. Again, it might be worth contacting iZotope. Perhaps they can offer some further suggestions. Best of luck!
8
Thank you very much for all your help! I actually had an odd amount of success with one portion I tried the other day, using the "remove vocals" function on adode audition. It seemed, oddly enough, to find most of the background chatter to be the vocals, and preserved most of the two voices and any background music. Albiet, they come out very quiet, but with a little amplification, im definately getting there. Thanks for all the help!
9
Quote:
im definately getting there. Thanks for all the help!

Glad to hear it! :bravo:
10
OK, So here's the deal. I'm having trouble planning on where to go here.

So after doing the Vocal Extractor (Central Channel Extractor), I end up with this. I left it raw and unamplified in case there is something I should do before I boost the volume. I used the default settings, and I messed with the settings an awful lot and ended up with different, yet more distorted versions. This seems to be the clearest voices, yet they are silenced to a whisper. There are a ton of artifacts in the result, but it would seem I can just delete them by highlighting all signals above a certain Hz and just deleting. I am having mixed results working with it, but it REALLY feels like I can get there with this method. Nearly the whole file almost silences the background and makes the voices more prevalent. Can anyone offer any ideas on this sample?

Again, this was the raw file, put through Adobe Audition with the Central Channel Extractor on the default "Vocal Extractor" settings, and not touched any further. Can anyone offer some guidance? Thanks again!

https://drive.google.com/open?id=0B5vTDz4l9lsWUlNPSXhYZTFwYWc