I have a phone recording of a conversation between 2 folks.
I want to generate 2 separate recording each having one's conversation part in it.
Is there a tool that can detect the 2 voices separates them into individual files?
Summary: Use a noise removal plugin that can learn noise to produce a track with one person "noisereduced" and use this track as sidechain source for a noise gate.
Here is how to achieve it with a noise removal plugin that can learn a sample, some EQ and a sidechainable noise gate:
Make two empty tracks and four copies of the track and name them like so:
Copy some different parts with person 1 to track A (order and cut noise is irrelevant - just make sure person 2 IS NOT present in these samples).
No there is not such a tool. Frequencies of human voice ranges from 300 to around 3500 Hz (in telephony). There are different amplitude of harmonics between a man, a woman or a child but it is not simple to separate the voices. Perhaps in modern voice recognition software there is such a capability, by tracking the "footprint" of each voice but i haven't seen an incarnation of this functionality for the task you are looking for. Manual separation with an audio editor is perhaps your only choice (as other posters pointed out).
Five years later and there are a number of cloud services that offer to solve this challenge. Services https://otter.ai/ and https://cloud.google.com/speech-to-text offer some alternatives for the problem of "speaker diarisation", which identifies when specific people are speaking. Once you have a list of times at which a specific person spoke in a recording, you would then isolate those sections of the recordings.
If you are indeed saying that the voices are speaking at the same time, and you have a SINGLE recording of both of them at the same time, AND it is in mono, it is not possible. You have to pre-emptively record each person's voice with two different microphones, recorders, what have you.
Since human beings naturally occupy essentially the entire frequency spectrum (strongest near 2000~10000 Hz), separating them via either EQ or panning without compromising their clarity is not possible unless they are on different channels or different recordings. Furthermore, on a phone recording (which tends to be in mono, and where there is probably a substantial lack of upper treble), it is much less likely that someone could do it via EQ or panning while also maintaining clarity since many syllables tend to reside in treble frequencies. My advice? Think ahead.
If the two people are speaking at different volume levels then you could make a gate that activates only when one voice crosses the threshold and vice versa. As it's been said before, it will be tough to separate the voices since they are in the same frequency range and share similar characteristics. You could set narrow band EQ's that could focus in on the desired voice but there would be overlap since the other voice most likely resides in the same frequency range.