Audacity can edit audio with AI through the free OpenVINO AI Tools plugin, which adds noise suppression, stem separation, transcription, music generation and audio upscaling to the editor you already use, and runs entirely on your own computer: no subscription, no upload, no internet connection. The plugin runs in Audacity 3.7 and does not load in Audacity 4 yet.
This guide covers what the five effects do and where each one sits in the menus, what to do if you already moved to Audacity 4, how to install the plugin through MuseHub, and the limits to know before you rely on it.
Tested on Audacity 3.7.9 on Windows 11 and macOS Sequoia, with the plugin installed through MuseHub; the screenshots are from the Windows install. The plugin runs in Audacity 3.7.4 to 3.7.9 on Windows and macOS. Audacity 4 does not load these effects: the plugin’s developers confirmed in September 2026 that the current version will not work in Audacity 4 and that the port is a large job with no date. If you are on Audacity 4, the next section covers what to do meanwhile.
AI tools for Audacity
The five AI effects at a glance:
- Cleaning up a voice recording: Noise Suppression, tuned for spoken word.
- Pulling a song apart: Music Separation, into drums, bass, vocals and everything else.
- Turning speech into text: Whisper Transcription.
- Rescuing a low-quality file: Audio Super Resolution.
- Generating something new: Music Generation and Continuation.
How we choose: we describe what each tool actually does and where it falls short, and we flag the cost of everything, free and paid alike.
What “AI editing” in Audacity actually means
Audacity on its own is a conventional audio editor: you cut, fade, EQ and export. The AI features are not built in. They arrive as a separate plugin pack, OpenVINO AI Plugins for Audacity (OpenVINO AI Tools on MuseHub), published by Muse Group and free under the GNU General Public License v3.
The name comes from OpenVINO, Intel’s toolkit for running machine-learning models on ordinary hardware. The models that normally live on a company’s servers run on your own processor instead (the CPU, or a supported GPU or NPU), and once installed they appear in Audacity’s menus alongside the classic effects and behave the same way: select some audio, choose the effect, set a few options, apply.
Running locally means no upload, so a client’s interview or an unreleased track never leaves your machine; no minute cap and no subscription, so a three-hour podcast costs the same to process as three minutes; and no watermark on the output. The trade-off is that processing runs at the speed of your computer rather than a server farm, so a long file takes real time.
Already on Audacity 4? Read this first
Audacity 4 shipped in September 2026 with a new plugin system, and the OpenVINO effects are written for the Audacity 3 one. The developers’ next release is still an Audacity 3 one, and the Audacity 4 port has no date. Until that changes you have two honest options.
Keep Audacity 3 installed next to Audacity 4. On Windows, the Audacity team says, the two versions install into different folders, C:\Program Files\Audacity 4 and C:\Program Files\Audacity, and keep separate preferences, so both can be open at once. Do the AI job in Audacity 3, export the result as WAV, then import it into Audacity 4 for the rest of the edit. Audacity 4 also opens Audacity 3 projects, so you can move the whole project instead.
Or wait, and use what Audacity 4 has. For noise, Audacity 4’s built-in Noise Reduction still works the classic way, and removing background noise in Audacity walks through it. For vocals and transcription, Audacity 4 has no built-in equivalent yet. What’s missing in Audacity 4 tracks when that changes. Separation does not have to wait, though: the free MuseHub Vocal Remover and Stem Splitter run in the browser and give you files to import into Audacity 4. The two deeper guides linked below cover the rest.
What you need before you start
Three things, and the first one causes almost every “the effects aren’t showing up” problem.
Audacity 3.7.4 to 3.7.9. The MuseHub listing states that range, and the plugin will not appear in older releases or in Audacity 4. Check Help ▸ About Audacity. Audacity 3 is free and installs from MuseHub under its own name, Audacity 3, separately from Audacity 4.
The OpenVINO AI Tools plugin, installed through MuseHub. It is the route the Audacity team points people to, and it handles a step that catches people out when they install by hand: during setup you choose which models you want. The models are the machine-learning files themselves, and they are large, so you are not forced to download all of them if you only want one tool.
A platform that has a build. MuseHub installs the plugin on Windows (64-bit) and on macOS, both Apple Silicon and Intel. On a Mac the effects run on the CPU only, so the heavy ones are slower: on our test Mac, the first four-stem Music Separation of a 30-second excerpt took about three minutes, against about ten seconds on Windows. On Linux there is no official installer: the community Audacity snap ships Audacity 3.7.9 with the plugin built in, and otherwise it is built from source, which is a developer job. MuseHub itself runs only on Windows and macOS. Windows on Arm is out: Audacity’s release notes say its Windows on Arm build supports no plugins, OpenVINO included.
You may see claims elsewhere that a specific Intel processor generation is required; neither the plugin’s documentation nor the Audacity pages state one. You pick the CPU, GPU or NPU in each effect’s dialog, and older hardware runs the effects more slowly, though very old processors can crash (see the troubleshooting section below).
The five AI effects
Since Audacity 3.7.4 all five OpenVINO effects have their own group in the Effect menu, OpenVINO AI Effects. That includes Whisper Transcription and Music Generation, which older guides put under Analyze and Generate.
Noise Suppression
The one most people install the plugin for. It strips background noise out of a recording, such as room hum, air conditioning or street noise behind a voice, and it is tuned for spoken word.
Find it under Effect ▸ OpenVINO AI Effects ▸ OpenVINO Noise Suppression. The dialog has an OpenVINO Inference Device selector, a Noise Suppression Model list and an Advanced Options checkbox. The models are DeepFilterNet 2 and DeepFilterNet 3 (deepfilternet3 in the list), which the developers recommend, and denseunet, kept for compatibility with older versions. There is a Preview button, but it stays greyed out. To judge the result, apply the effect to a short, typical stretch, listen, and press Ctrl+Z (⌘Z on a Mac) to undo if it went too far. The first run on a device adds ten to thirty seconds while the model compiles; later runs skip that.
Unlike Audacity’s classic Noise Reduction effect, it needs no noise profile: there is no half-second of silence to find and sample first, which matters on recordings that never have a clean moment of just-noise.
Best for: interviews, podcasts, voiceovers, video dialogue.
Music Separation
Splits a finished stereo mix back into its parts. Under Effect ▸ OpenVINO AI Effects ▸ OpenVINO Music Separation you pick a mode, “(2 Stem) Instrumental, Vocals” or “(4 Stem) Drums, Bass, Vocals, Others”, and the Inference Device; there is also an Advanced Options checkbox. Each stem lands as a new track below the original, named after it with the stem appended (-Drums, -Bass, -Other Instruments and -Vocals in 4-stem mode), so the original stays untouched. A 4-stem run on a 30-second excerpt took about ten seconds on the test machine’s CPU. The model is Meta’s Demucs, ported to run through OpenVINO.
Separation is reconstruction rather than un-mixing: a dense, heavily compressed master leaves a watery shimmer around the vocal and cymbals bleeding into the Other Instruments stem, while a sparse arrangement can come out startlingly clean. The full walkthrough, with the built-in Audacity 4 route and the stem-separation plugins on MuseHub in one table, is in how to remove vocals in Audacity.
Best for: karaoke and backing tracks, remixing, sampling, transcription practice.
Whisper Transcription
Converts speech to text using Whisper, OpenAI’s speech-recognition model. It sits with the other OpenVINO effects under Effect ▸ OpenVINO AI Effects ▸ OpenVINO Whisper Transcription, even though it produces a label track rather than changing your audio. You pick a model (base, small, medium or large, in increasing accuracy and processing time) and a mode: transcribe keeps the source language, translate outputs English whatever the source. The words land as labels anchored to the moments they were spoken, and File ▸ Export Other ▸ Export Labels writes them out as plain text, SRT or WebVTT subtitles, or podcast chapters in JSON.
Model choice, subtitle export, the accuracy limits and the alternatives are in how to transcribe audio in Audacity.
Best for: podcast show notes, subtitles, interview logs, finding the one sentence you need in a two-hour recording.
Audio Super Resolution
Upsamples audio to improve its clarity: whatever you select is rebuilt at 48 kHz with the top end filled in. The clearest use case is old, low-bandwidth material such as 8 kHz telephone recordings, where the top end was never captured. Under Effect ▸ OpenVINO AI Effects ▸ OpenVINO Super Resolution you choose a Model (General for music and mixed material, Speech for voice) and the device, with a Normalize Output RMS option that keeps the loudness where it was.
It cannot invent detail that was never there, and it will not turn a phone recording into a studio session. What it does is make thin, muffled archive material noticeably easier to listen to.
Best for: archive tape, old voice memos, phone and video-call recordings.
Music Generation and Continuation
Generates music from a text prompt, or continues an existing clip, using Meta’s MusicGen. It sits with the other four under Effect ▸ OpenVINO AI Effects ▸ OpenVINO Music Generation, even though it creates audio rather than changing it, and takes a Prompt, a Duration, a model (mono or stereo, at 32 kHz), a Seed for repeatable results and a Guidance Scale that sets how closely the output follows the prompt. The continuation half is the more useful one: select a few bars and let it extend them to stretch a piece of underscore to fit a video.
Best for: rough ideas, filling out an underscore, experimenting.
The five effects compared
| Effect | What it does | Where to find it | Best for |
|---|---|---|---|
| Noise Suppression | Removes background noise from speech | Effect ▸ OpenVINO AI Effects | Podcasts, interviews, voiceover |
| Music Separation | Splits a mix into 2 or 4 stems | Effect ▸ OpenVINO AI Effects | Karaoke, remixing, sampling |
| Whisper Transcription | Speech to text as a label track | Effect ▸ OpenVINO AI Effects | Show notes, subtitles, logging |
| Audio Super Resolution | Upsamples low-quality audio to 48 kHz | Effect ▸ OpenVINO AI Effects | Archive and phone recordings |
| Music Generation | Creates or extends music from a prompt | Effect ▸ OpenVINO AI Effects | Sketching, underscore |
How to install it
The straightforward route is through MuseHub, the free app that installs and updates Audacity plugins and instruments. MuseHub needs a free account and runs on Windows and macOS.
- Install or update Audacity 3 from the Apps section in MuseHub (it is listed separately from Audacity 4) and confirm the version is 3.7.4 to 3.7.9 under
Help ▸ About Audacity. This is the step to get right first. - Open OpenVINO AI Tools in the Plugins section and start the install. On Windows, MuseHub launches the plugin’s own installer, titled Setup - OpenVINO AI Plugins for Audacity.
- Choose the models. The installer asks for a destination folder and for the models; it does not ask which effects you want. With Install recommended models, it ticks Whisper Base, Music Separation, Super Resolution, DeepFilterNet 2 and 3 for Noise Suppression and Music Generation Small Mono, at least 4.04 GB in total. The larger Whisper models (Small, Medium, Large v1 to v3) are extra ticks. Only take what you will use; the model files are large and take time to download.
- Restart Audacity 3. All five effects appear in the Effect menu under OpenVINO AI Effects.
On a Mac, MuseHub opens a Model Downloader window instead, with the same model packs. It preselects one Whisper model, about 171 MB, and nothing in Noise Suppression or Music Separation, so tick those if you want them; its Whisper list starts at Small.
There is also a manual route: the installers are linked from the Audacity team’s OpenVINO download page. Close Audacity first, and on Windows point the installer at the Audacity 3 folder if you have both versions, because it may pick the wrong one.
OpenVINO effects missing or crashing
Work down this list in order; most cases end at one of the first three items.
- You opened Audacity 4. The effects do not load there, because Audacity 4 has no way to load modules such as OpenVINO yet. The Audacity team says support is planned but has not announced a date. Open Audacity 3 for the AI jobs.
- The plugin build does not match Audacity. Each OpenVINO build is made for a specific Audacity 3 version, and a mismatch can crash Audacity at launch. Installing through MuseHub picks the matching build; after a manual install, compare the two version numbers.
- You are looking under Analyze or Generate. Since Audacity 3.7.4, Whisper Transcription and Music Generation sit with the other OpenVINO effects under Effect ▸ OpenVINO AI Effects.
- An effect fails and the Inference Device list is empty. After a reinstall from MuseHub with Audacity open, Music Separation stopped with “Music Separation failed. See details in Help->Diagnostics->Show Log…”. Restarting Audacity brought CPU back into the list.
- The effects are missing after a manual install. A MuseHub install sets the plugin’s module, mod-openvino, to Enabled on its own, so this is mostly a manual-install problem. Open Modules preferences (Edit ▸ Preferences ▸ Modules on Windows, Audacity ▸ Preferences ▸ Modules on a Mac), set mod-openvino to Enabled, press OK and restart Audacity. Still missing? Open Effect ▸ Plugin Manager and click Rescan.
- An error names a missing model file, such as
htdemucs_v4.xmlfor Music Separation. That model was never downloaded, because the installer fetches only the models you ticked. Run the plugin’s installer again without uninstalling first, and tick only the models you are missing. On Linux, the snap package downloads its models withsudo audacity.fetch-models --batch. - Audacity crashes on a much older computer. Forum reports include a crash on a processor from around 2011. Neither Intel nor the Audacity team publishes a minimum processor for the plugin, so if the effects crash on hardware that old, use the built-in effects instead.
Limits to know
Processing takes real time. These models run on your hardware, and there is nothing to pay to speed it up.
Noise Suppression is tuned for speech. On music it can do odd things to sustained instruments and cymbals. For a vinyl rip or a music recording, Audacity’s classic Noise Reduction, or a dedicated plugin, is usually the better call.
No Audacity 4 support, and no official Linux installer. In Audacity 4 there are no OpenVINO effects until the developers finish the port, on Windows and on a Mac alike. On Linux it means the community snap or a build from source.
The models take disk space, which is why the installer asks which ones you want.
Audio in the screenshots: the song in the Effect menu and Music Separation shots is “Pop Vocal” by The_Mountain on Pixabay, used under the Pixabay Content License.
FAQ
Is the OpenVINO plugin free? Yes. The plugin is free and open source under the GPL v3, and Audacity itself is free. There is no subscription and no per-minute charge.
Does OpenVINO work with Audacity 4? Not yet. The current release runs in Audacity 3.7.4 to 3.7.9 only; the developers confirmed in September 2026 that it will not load in Audacity 4 and that the port has no date. Keep Audacity 3 installed alongside Audacity 4 for the AI effects, or wait.
Why can’t I see the AI effects in Audacity?
Almost always the Audacity version. Check Help ▸ About Audacity: you need 3.7.4 to 3.7.9, and if Audacity 4 is also installed, make sure you opened Audacity 3. All five effects sit under Effect ▸ OpenVINO AI Effects, not Analyze or Generate. The troubleshooting list above covers mismatched builds, disabled modules, missing models and an empty device list.
Does my audio get uploaded anywhere? No. Everything runs locally on your computer. Nothing is sent to a server, which is the main reason to use these rather than a web-based tool for confidential material.
Does it work on Mac? Yes, in Audacity 3.7.4 to 3.7.9, installed through MuseHub like on Windows. The effects run on the CPU, so Music Separation is much slower than on a Windows PC. In Audacity 4 the plugin does not load on any platform yet; for stems there, the free MuseHub Vocal Remover and Stem Splitter do the separation in the browser.
Can it remove noise from music, not just speech? Noise Suppression is tuned for spoken word. It will run on music, but results are unpredictable on sustained notes and cymbals; use the classic Noise Reduction effect or a dedicated plugin instead.
Next steps
The two AI jobs people ask about most have their own guides: how to remove vocals in Audacity compares Music Separation with the built-in route and the stem plugins on MuseHub, and how to transcribe audio in Audacity takes Whisper from model choice to exported subtitles.
For everything else you install, how to use effects and plugins in Audacity covers both ways of applying an effect in Audacity 4, and the best Audacity plugins lists what is worth adding, with the formats Audacity 4 loads. Removing background noise in Audacity covers both noise routes, and the best AI plugins covers AI tools that work inside a DAW rather than in Audacity.
Get OpenVINO AI Tools, Audacity 3 and the rest of the Audacity plugin catalogue in one place — download MuseHub free.
