Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

iMas70

macrumors 65816
Original poster
Sep 4, 2012
1,440
206
MA
I recently used my iPhone to record a conversation but there was music playing close by. I'd like to be able to take that out so I can clearly hear the conversation. What would be a good program to use for this?

Thanks!
 
  • Like
Reactions: rm5

rm5

macrumors 68020
Mar 4, 2022
2,407
2,753
United States
I recently used my iPhone to record a conversation but there was music playing close by. I'd like to be able to take that out so I can clearly hear the conversation. What would be a good program to use for this?

Thanks!
It depends on how loud the music is vs. how loud the speaking is. The Audacity noise reduction tool will remove noise using a gate, so it cuts it out based on a set decibel value. If the difference between the decibel value of the music and the speaking is small, some of the music will remain. If the difference is large, you'll have no problem.

I'll post an example of what I'm talking about later, it's hard to explain.
 
  • Love
Reactions: KaliYoni

ignatius345

macrumors 604
Aug 20, 2015
7,102
11,718
At this point there may also be some AI-based tools that could apply some logic to this and do what you need -- assuming you're comfortable (probably) having to upload your file to something to have it processed.
 

rm5

macrumors 68020
Mar 4, 2022
2,407
2,753
United States
Okay, here's an example (don't laugh at the cringy drum beat lol). I read a short part of a script as an example.


You can hear that with the noise reduction, the audio REALLY suffers. I set it to reduce by 48 dB, which is the maximum amount. If you listen super close, you can hear small remnants of the drum beat in the background, even on the noise reduced recording, so it certainly doesn't get rid of everything (at least in this case).

lmk if you want more specific recordings with different settings
 
Last edited:

Slartibart

macrumors 68030
Aug 19, 2020
2,928
2,663
Well, depending on the recording you can e.g. try to remove the vocals first using Audacity’s Nyquist plug-in (via Effects>Vocal extraction and Isolation)and then capture the noise profile of the resulting track that has had vocals removed, and run Noise Reduction with that profile on the original recording before vocals were removed.
This might give a better result.

There is a detailed description in the Audacity manual, please take a look on Case 3.
 
  • Like
Reactions: CooperBox

iMas70

macrumors 65816
Original poster
Sep 4, 2012
1,440
206
MA
Thanks for the info and examples! I'll check this one out. The music in my recording drowns out the person who is speaking sometimes. I don't mind if it's not totally gone but I'd like to be able to hear what was being said.
 
  • Like
Reactions: rm5

iMas70

macrumors 65816
Original poster
Sep 4, 2012
1,440
206
MA
Okay, here's an example (don't laugh at the cringy drum beat lol). I read a short part of a script as an example.


You can hear that with the noise reduction, the audio REALLY suffers. I set it to reduce by 48 dB, which is the maximum amount. If you listen super close, you can hear small remnants of the drum beat in the background, even on the noise reduced recording, so it certainly doesn't get rid of everything (at least in this case).

lmk if you want more specific recordings with different settings

I've been working on this for about an hour. The music is very prominent. It's pretty much in the foreground with the person speaking in the back. With the noise reduction tool, besides reducing the music, it also reduces the volume of the person speaking. I tried using the vocal extraction and isolation tool. That only works with stereo tracks.

I haven't really worked with stuff like this much. Just edited some audio to go along with video in the past. What I was picturing is something that shows the music track and the other sounds like the person speaking in the background. Then be able to cut out the music to just leave the background conversation.
 
Last edited:

rm5

macrumors 68020
Mar 4, 2022
2,407
2,753
United States
I've been working on this for about an hour. The music is very prominent. It's pretty much in the foreground with the person speaking in the back. With the noise reduction tool, besides reducing the music, it also reduces the volume of the person speaking. I tried using the vocal extraction and isolation tool. That only works with stereo tracks.
That's the one drawback of this feature, is that you lose a LOT of the speech, too, along with the music. Without some AI tool (that you'd probably have to pay for), I think doing it in Audacity is about the best you're gonna get, unfortunately.
 

Slartibart

macrumors 68030
Aug 19, 2020
2,928
2,663
You have the music/instrumental track?
  • In that case you load the “contaminated” and the instrumental track both into audacity.
  • Select the Time Shift Tool to roughly align the two tracks properly. Next, zoom in, you want to see the wave function very closely. Take the proper time to align this as closely as you can, e.g. pick a peak or trough in the left channel of one track and match it precisely with the left channel of the other track. If the alignment isn’t right, the process won’t really work.
  • Switch to the Selection Tool. Double click on the instrumental waveform to highlight all of just that one track, and go to Effect > Invert.
  • Then use +A to select all of both tracks. Then use Tracks > Mix and Render.
You’ll get one combined track that should have a more diminished amplitude where the vocals were kept and the instrumentation removed. The important bit here is the alignment. 🤓
From there you can improve the audio track you obtained further.

EDIT: the better the quality of the original tracks, the better the effect works. If you’re using MP3s, avoid using tracks with a lot of clipping. This will ruin the effect over those parts. You can highlight where clipping occurs on your tracks in Audacity via View > Show Clipping.
 
Last edited:
  • Like
Reactions: ignatius345

traveltoromantis

macrumors regular
Oct 19, 2017
102
110
You have the music/instrumental track?
  • In that case you load the “contaminated” and the instrumental track both into audacity.
  • Select the Time Shift Tool to roughly align the two tracks properly. Next, zoom in, you want to see the wave function very closely. Take the proper time to align this as closely as you can, e.g. pick a peak or trough in the left channel of one track and match it precisely with the left channel of the other track. If the alignment isn’t right, the process won’t really work.
  • Switch to the Selection Tool. Double click on the instrumental waveform to highlight all of just that one track, and go to Effect > Invert.
  • Then use +A to select all of both tracks. Then use Tracks > Mix and Render.
You’ll get one combined track that should have a more diminished amplitude where the vocals were kept and the instrumentation removed. The important bit here is the alignment. 🤓
From here you can improve the audio track you obtained further.

EDIT: the better the quality of the original tracks, the better the effect works. If you’re using MP3s, do whatever you can to avoid using tracks with a lot of clipping. This will ruin the effect over those parts. You can highlight where clipping occurs on your tracks in Audacity by going to View > Show Clipping.
In my experience, this doesn't really work very well unless the two tracks are pretty much exactly the same quality or very similar, which in this case it wouldn't be. For a similar but more complicated method that does work with tracks of differing quality, I'd reccomend using the Kn0ck0ut plug-in in Audacity. It's an old plug-in but it's still worth looking into.

The easiest way to use the plugin would be to:

  • Follow the first two steps from Slartibart's post.
  • Convert both tracks to mono ('stereo track to mono').
  • If the tracks are at significantly different volume levels, I would advise adjusting them as close as possible with the Amplify tool.
  • Set the conversation track to the left channel, and the song to the right channel.
  • Mix/render the tracks to a single stereo track.
  • Select the new stereo track and run kn0ck0ut. Adjust the R Input gain slider within the plugin for different results.
 
Last edited:
  • Like
Reactions: Slartibart

iMas70

macrumors 65816
Original poster
Sep 4, 2012
1,440
206
MA
I don't have the tracks for the music. That was random songs that was playing.
 

bogdanw

macrumors 603
Mar 10, 2009
5,751
2,773
I don't have the tracks for the music. That was random songs that was playing.
You don't need tracks. When you first run spleeter separate -p spleeter:2stems -o output the model "Vocals (singing voice) / accompaniment separation (2 stems)" will be downloaded and your file processed.
 
  • Like
Reactions: Slartibart

rm5

macrumors 68020
Mar 4, 2022
2,407
2,753
United States
I've had very good luck using DaVinci Resolve to isolate voices.
I've heard really great things about the voice isolation tool in Resolve as well... but doesn't that require the Studio version though? Unless you already have it, it's probably not worth buying it solely for the voice isolation tool.
 

Bananasaurus

Suspended
Aug 16, 2023
753
2,631
I've heard really great things about the voice isolation tool in Resolve as well... but doesn't that require the Studio version though? Unless you already have it, it's probably not worth buying it solely for the voice isolation tool.
Yeah, sorry. I forgot it's only in the Studio version. I've moved away from Final Cut and Premiere to DaVinci over the last year and I'm not going back.
 
  • Like
Reactions: rm5
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.