I have an audio file of a man talking and some conversation taking place in the background. I need to remove the main voice and emphasize the background conversation. Someone can help me?
There is no generic recipe to remove a certain specific sound from a recording. It may or (most probably) may not be possible to do what you wish, as voices are voices and take more or less the same place in the frequency spectrum, so there is no "magical" way of isolating them. Depending on the specific case, it may me possible to attenuate the voice you don't want, probably at the expense of the overall quality and perceptibility of the rest. So it's a matter of experimentation and trying to get the best compromise.
It's hard to summarize an approach without going into lengthy technical explanations, and to be honest, I think it will be extremely difficult for someone without the proper training to do it, but even so it may be worthwhile to lay out the basic approach.
Basically, you need to identify frequencies in your recording that belong to the voice (ot other sounds) you want to eliminate, but not (as much as possible) to the voices you want to keep. Then you need to apply filters to remove these frequencies from the audio. Best results (or perhaps, not so bad results) may be achieved if this process is done specifically to different audio segments, depending on how the voices mix along the recording.
In practical terms I would suggest not to start with Audacity (as excellent tool as it is), but to a more specialized audio analisys tool. I recomend Sonic Visualizer as a free/open source tool widely used for audio analisys in research and academia (setting up Sonic Visualizer with the required plugins may be a bit of a project on it self, but there are instructions in the site and it's worth the trouble).
Load your audio into Sonic Analiser and perform a spectral analisys. The details of the resulting spectrogram will intirely depend on the specific audio, but here's an example of a spectrogram (probably from a piece of music):
Listen to your audio while looking at the spectrogram and see if you can visually correlate the sounds you want to eliminate with specific high energy presence in the spectrogram (i.e. the red lines). If so, find on the left hand scale the frequency (or frequency band) where it seems to fit mostly. Take notes of the time intervals and frequencies.
Then go to Audacity and apply Notch Filters for each time interval and frequency you previously identified. You'll probably need to repeat the process multiple times, refining the parameters. (There are probably tools that allow the analysis and filtering steps in the same tool, but I don't know them and, unless I'm mistaken, they are not publicly and/or freely available anyway)
Be prepared for the outcome to be that it is practically impossible. You don't mention WHY you want to do this, or WHAT kind of result you need. Are you creating a sound-track for a video? Are you trying to understand what someone in the background is saying? Do you need "broadcast quality" Do you need "transcription quality"? It is pretty unlikely that you can accomplish this with Audacity. There are software tools that cost as much as your car which are used by forensic professionals for applications like this.
ESSENTIALLY, -you need to find the frequency that the main voice is talking in -you need to find the frequency that the background voices are talking in -increase dynamics of background frequencoes -decrease dynamics of main frequencies
This is incredibly difficult as those frequencies will overlap and you will most likely be altering a RANGE of frequencies
There are probably plugins that can help you out a little bit but I would say this is mostly up to you having patience and taking this one step at a time.
There is no technology to separate dialogue from background dialogue sounding at the same time. Especially while the background dialog sounds lower than the main speaker. You better forget it and don't waste your time. The only thing you can do is to try to figure out what's been said in the recording as it is by adjusting your overall volume if need it.
I know this is a ¾-decade-old question and the person has likely given up by now, but there is a commonly used program that can raise background voices to at least come close to matching foreground voices, despite not being a forensic application. IIRC it was even able to do this back in 2016 when this question was asked.
The program is Adobe® Audition® and has gone through many versions. It was stand-alone for awhile, then part of Creative Suite® (Audition CS#), then Creative Cloud® (CC[year] until recent years, and now just [year]).
The trick is a built-in filter (no need to download any VSTs, etc.) that’s not in an obvious place in most versions. In the latest Adobe Audition 2024, it’s under “Effects” ⇒ “Amplitude and Compression”. In earlier versions it may’ve been in different groupings including an Obsolete Filters grouping.
It’s called “Speech Volume Leveler…,” and it does exactly what its name implies: it levels the volume of speech in audio. The dialog gives you considerable control over how much leveling it does. It’s available in the Effects Rack of newer versions, too, which means you can use it in real time, non-destructively, if your computer’s fast enough.