AI-powered “Enhance Speech” is able to make lower-quality audio recordings sound professional.
Recently, Adobe unveiled a free AI-powered audio refining tool that can improve poor-quality voice recordings by reducing background noise and enhancing the sound of the voice. When operational, the output resembles a professionally recorded sound with a top-notch microphone.
Formerly known as “Project Shasta”, Adobe’s new AI tool to help improve audio recordings is now called “Adobe Podcast”. This project was initially created for research purposes.
Enhance Speech is free to use, but an Adobe account needs to be created and a desktop web browser works best.
Upon registration, users can upload an audio file with a maximum length of 1 hour or 1GB. After a few moments, the output can listen to online or downloaded as a cleaned-up version.
What Is It Most Effective With?
According to our tests, Enhance Speech was most effective when audio had a clean voice without outside noise or interference.
For instance, after processing the audio (using Enhance Speech) recorded from an iMac’s built-in microphone with a person standing 10 feet away and a fan nearby, it sounded as if it had been recorded in a studio up close with a professional mic in silence.
How Does It Work?
Without giving any further specifics, Adobe probably trained a deep-learning model on a large set of audio files; both clean and with noise. This model was likely able to discern the frequencies specific to human voices, thus making it possible to synthesize audio that accurately resembled the source.
Until Adobe provides more technical information, this is mere speculation and we have reached out to the company for further details.
Some Hacker News commenters have mentioned that Enhance Speech produces hallucinated results—unexpected output like phantom voices when it interprets audio—from sources with extreme noise (such as an audio recording beside a waterfall) and non-English languages.
This implies that Enhance Speech is doing more than enhancing sounds through noise reduction.
Enhance Speech isn’t the only service to offer AI-powered noise relief technology; mayavoz, an open-source package, and Audio Studio, a commercial service, have similar capabilities.
It should be mentioned that Enhance Speech is part of Adobe’s larger group of AI-powered podcasting tools, including a Mic Check tool (which is free to use at the moment) and a transcript-based audio-editing tool which is currently in its invitation-only beta testing phase.