Adobe Podcast Enhance alternatives, compared on the same recordings

Is Adobe Podcast Enhance generative AI?

Yes. Adobe Podcast Enhance, also called Enhance Speech, produces a new performance of your voice rather than filtering the one you recorded. A filter can only make a sound quieter; it cannot add one that was never there. In Premiere, Enhance Speech has been reported adding in random words, hallucinating voices during gaps in speech and changing “Park” into “farch”. On the web, it has been reported changing a speaker’s accent and turning Spanish into an unrecognizable language. Only a model that produces speech can do any of that.

It is also how Enhance Speech takes a whole room out of a recording, which subtraction never managed. The catch is that the result is only as good as the model’s rebuild of what it could not hear: the words, the accent, and the sound it puts around them.

How does Adobe Enhance Speech work?

Adobe Enhance Speech works like any regenerative enhancer: it works out what was said and what the voice sounds like, then produces that speech again with no room and no noise in it. Adobe does not publish its internals, so this describes the kind of tool it behaves like. Where the recording is clear, you hear the room gone. Where it is not, the model still has to produce something, and it produces the most likely speech: usually what you said, sometimes a word that sounds close, a more average accent, or a voice in a pause where nobody spoke.

A filter that goes wrong sounds wrong everywhere. A model that guesses sounds right everywhere, including where it is wrong. The texture side of the same trade, a voice that comes back too smooth or glassy, is in why AI-enhanced audio sounds robotic.

Adobe Podcast vs ElevenLabs vs Crele, on the same recordings

On the same recordings, Adobe Enhance Speech V2 lets background noise leak through and gives the voice a metallic edge, ElevenLabs Voice Isolator adds small hallucinations and hands back a bare voice, and Crele Remaster 1 takes the noise out and makes the voice sound recorded in a studio. Hear it for yourself: two recordings, each through all three, cut to the same stretch and set to the same loudness so none of them wins by being louder.

Each tool has its own sound, and you can hear it on both recordings:

Crele Remaster 1 rebuilds the voice, like Adobe’s, and aims for what a studio would have given you: the noise gone, the words as they were said, and the sound of a close microphone in a treated room.

How to remove a TV or another voice from a recording

To remove a second voice, from a TV, a co-host or mic bleed, you need speaker isolation, not an enhancer. An enhancer hears every voice as speech and keeps it; as an answer on Adobe’s own forum puts it, Enhance Speech “doesn’t know who ‘Speaker A’ or ‘Speaker B’ is.”

Crele Focus 1 does. It learns one voice from a few seconds of that person talking alone, follows it through the recording, and removes everyone else. Run it first, then enhance the one voice it keeps with Crele Remaster 1.

The longer version is in why noise removal can’t remove a second voice.

How to evaluate a speech enhancer

To evaluate a speech enhancer, run the same untouched recording through every tool, match the loudness, and listen blind, back to back. Every tool sounds good on its own demo; these are the checks that show the difference on your audio:

• Use your worst recording, untouched. Differences show on a noisy, distant or phone recording, and only if every tool gets the original file, not one another tool has already cleaned.
• Match the loudness first. The louder of two versions always sounds better, and every tool hands back its own level. Set them to the same loudness before you judge, or you are picking the loudest.
• Listen blind, back to back. Have someone else play the versions without saying which is which, and switch between them on the same sentence. Knowing the tool changes what you hear, and switching straight from one to the next is what makes a small difference obvious.
• Check the words against the original. Names, numbers and places are where a rebuilding model guesses, and a changed word sounds perfectly natural. Listen alongside the original, not to what you expect to hear.
• Listen in the pauses. That is where noise leaks through, and where a voice can appear from nothing.
• Listen for other voices. A TV or a second person is speech too, and an enhancer may clean it up instead of taking it out.
• Then listen to the voice itself. Metallic, thin or bare are the common faults. What you want is the voice as a close microphone in a treated room would have caught it.

Questions

What is a good alternative to Adobe Podcast Enhance?

If what comes back from Enhance leaks background noise, sounds metallic or no longer sounds like a real recording, try Crele Remaster 1. It rebuilds one voice from an everyday recording, such as a phone voice memo, a call or a noisy room, and hands it back sounding as if it had been recorded on a close microphone in a treated room. It runs on a free Crele account.

Is Adobe Podcast Enhance generative AI?

Yes. A filter can only take sound away, and Enhance Speech has been reported adding words that were never said, which only a model that produces speech can do. What it hands back is a new performance of your voice: that is how it can take a whole room out of a recording, and why it can get a word wrong.

How is Crele Remaster 1 different from Adobe Enhance Speech?

Both rebuild the voice rather than filtering the recording, so the difference is in what comes back. On the recordings on this page, Adobe lets background noise leak through and gives the voice a metallic edge, while Crele Remaster 1 takes the noise out and gives the voice the ambience of a studio recording. When another voice is in the file, Crele Focus 1 keeps the one you want and removes the rest, a job an enhancer is not built for.

Can I remove a TV or another person from a recording?

Yes, with speaker isolation rather than enhancement. An enhancer keeps every voice, because a TV or a second person is speech too. Crele Focus 1 learns the voice you want from a few seconds of it alone, follows it through the recording and removes everyone else, the TV included. Run it first, then enhance the one voice it keeps with Crele Remaster 1.

Does Crele Remaster 1 change words?

It rebuilds the voice too, and it is built to keep what you said and give the voice the sound of a studio recording. On a take with very little voice in it, any model that rebuilds speech is guessing, ours included. The comparisons on this page put it through the same recordings as Adobe and ElevenLabs, so you can hear the difference yourself.

Hear your own recording rebuilt

Make a free Crele account and run your own recording through Crele Remaster 1. Then compare it with the original the way this page does: at the same volume, word by word, in the pauses, and for the sound of the voice itself.

If a TV or another person is in the file, start with Crele Focus 1 to keep the voice you want, then enhance it.

Related: why AI-enhanced audio sounds robotic · why your recording sounds like a bathroom · why noise removal can’t remove a second voice