OpenVINO Tools Review by John Gieske

With hundreds of new AI tools available, it can be hard to know which ones to even try. But the free OpenVINO suite of AI tools integrates directly into Audacity, making them a logical addition to your workflow. The five tools are Noise Suppression, Music Separation, Whisper Transcription, Super Resolution, and Music Generation. Let’s look at each tool and how it could be useful to us.

Noise Suppression

Didn’t hear that cloth rustling during the recording ? Run Noise Suppression over the word that was impacted, and it’s gone. I spend less time in spectrogram view now that I can let AI find the noise. Noise suppression has a strength setting, from 1-100. Setting it to 100 will remove all noise but will also impact the voice, so I generally set it to about 20. Noise Suppression isn’t perfect. It especially struggles with short sounds that cover most of the frequency range (like a sharp bang). Nevertheless it does an impressive job of eliminating most background noise and can even reduce some echo. However it’s a resource hog, so I wouldn’t use it as a substitute for the classic Noise Reduction tool. Instead, use this for the occasional sounds that sneak in. (You can listen to some examples here)

Music Separation

Last year I recorded three drums, two pianos, two guitars, and half a dozen singers in an empty church. The result was a muddy mess. OpenVINO Music Separation rescued these songs. They still weren’t perfect but they were much better than they would have been without this tool. I now use Music Separation for many things: sometimes to pull guitar out of a lead vocal, sometimes to remove vocals from instrument tracks (which gives the vocals more presence), sometimes to change the sound of an obnoxious piano, sometimes to reduce the drums bleeding into vocals. Separating your music gives you much more control over the final sound. (See an example here)

My favorite use of Music Separation was removing the lyrics from an AI song that I had created and replacing them with my own in order to create a song for a volleyball team. You could similarly do a cover of a foreign worship song, changing the words into a local language while keeping the original instruments.*

Another use of Music Separation would be to dub films that are not in the IMS catalogue.* Use Music Separation to remove the voice from the background music and effects, then replace the voice with another language while keeping the original music. This isn’t perfect, especially because the tool isn’t made to handle sound effects, but for a quick project it would do a reasonable job.

Whisper Transcription

This tool transcribes your audio onto a label track, with each label corresponding to a pause in your speech. In English, it transcribed correctly even when I was muttering or mispronouncing words. It also works in 100 other major languages. It was only about 90% accurate when I transcribed my French, though I’m sure my poor accent was somewhat to blame.

As another test, I transcribed a recording into Hausa. The audio was a West African language unrelated to Hausa, but the transcription was close enough that I could see where the audio recording and the printed script lined up. So if you need to find one phrase in a long chapter, Whisper Transcription could save you from listening through the entire recording to find it.

Super Resolution

I have never used this tool because I always record 24-Bit/48kHz WAVs. But some day, someone will send me such a compressed recording that I’ll be afraid to edit it. On that day, I will be glad to have this AI upsampling tool. It has many parameters to tweak, it’s slow, and the results are mixed. Think of it as a last-ditch rescue plan rather than a regular part of your workflow.

Music Generation

Music Generation is underwhelming compared to similar AI tools. After taking a very long time to process, it can produce a dry and repetitive 20-second instrument track that nods in the general direction of the prompt you give it. Since it’s a 3 GB download and doesn’t work very well, I would recommend skipping it. (Samples of these might hurt your ears.)

For more information about the OpenVINO tools, visit the Github page here. You can download the Windows installer from here, which allows you to choose which OpenVINO tools to install. If you select everything, it’s about 7 GB; so make sure that you are on a good internet connection. The OpenVINO tools also work on Linux and Mac.

If you hunt around you can probably find AI tools that work even better than these ones, but the advantage to the OpenVINO tools is that they are available right there in the Audacity Effect menu (or by shortcut!). For my work, they have been a welcome additional to my toolbox. Try them out and see what you think!

*With permission, of course

John is a Vernacular Media Consultant who lives in Senegal with his wife, Sarah, and their four children. He has a BA in Global Mass Communications and 15 years of experience on the field, especially in audio recording and production. He often has the privilege of accompanying audiovisual projects from the early planning stages, all of the way through to distribution and promotion. John is passionate about training others and especially wants to see local believers producing scriptural audiovisual content for themselves.