Soften Harsh Voice In a song


I’m using Mac OS 12.6 and Audacity 2.4.2

I recorded a song of a friend in 1970 on a multitrack reel-to-reel, this was digitized to MP3 by a friend in 2016.

The guitar track and vocals are now combined into a stereo format, and each of the left and right channels are identical.

The vocals seem harsh to me and I want to soften these, basically, I want to reduce the level of nerve-twanging intensity they induce.

I tried noise reduction of the harshest sections of the vocals but they still sound sharp but basically slightly quieter.

I also tried “Sliding Stretch” to transpose to 432 Hz also without the desired result.

Is there a way to soften the voice on these recordings so they are not so harsh sounding? Any ideas are welcome.

Thank you.

It’s possible to manually pick out most of the gritty bits using Audacity’s spectral editing

But it would take many hours to manually clean a 3 minute pop song in that way.
The new AI cleaners are much faster, but require powerful hardware.

I appreciate the reply Trebor, thank you.

In the meantime, I found an online program that can separate the guitar from the voice, and this seems to have done a satisfactory job to separate them.

I’m a complete amateur with Audacity, but I played around with noise reduction and filter curves at random for the now isolated vocals track; without really knowing what I was doing.

I watch a few videos, but these were mainly about narration and how to improve podcaster voices, the instructions there didn’t take away the harsh tone of the vocals

The vocals are still harsh across most of the song, is there any filter to reduce the harshness, even carpet bombing to obtain any slight improvement?

I understand your comments on new AI cleaners.

Just any improvements at all would be better than the current quality like a one size fits all quick fix?

I found an online program that can separate the guitar from the voice

And what might that online program be? This is a forum, with users helping each other, not a help desk.


A quick & dirty method would be to increase the contrast of the spectrogram,
which reduces the velcro-like noises, then drown it in reverb, (which can hide a multitude of sins) …

Increasing the contrast adds digital artefacts, (but so will using spleeter or similar separation software).

The contrast-enhancing plugin I used was an old one called DtBlkFx
The reverb I used was Valhalla supermassive, but any reverb would do.

To do justice to your 70’s recording would be like restoring a painting: it would take weeks to do it properly.

Thank you for the great help Trebor, I appreciate your time and input.

The samples you attached are certainly better and the improvement is sufficient for what I wish to achieve.

Now I have the vocals separated out from the guitar the changes you recommend should be the solution.

Kind regards.