Audacity Stem Separation in 2026: AI Stem Splitter, OpenVINO Plugin & Vocal Reduction Compared

Audacity is the most-used free audio editor in the world — and as of Audacity 3.5 (May 2024), it has a real, AI-powered stem splitter built in: the OpenVINO Music Separation plugin. That changes the game for free vocal removal if you're on Windows. For Mac, Linux, and ChromeOS users, the answer is different.

The short answer:

• For the cleanest stems on any operating system → external AI (StemSplit) at $0.10/min — works in any browser, drag the WAVs straight into Audacity.
• For free local AI on Windows → install the OpenVINO Music Separation plugin from audacityteam.org/download/openvino. Uses Demucs v4 (htdemucs). Free, runs offline, takes 30s–5min depending on your CPU/GPU.
• For Audacity's built-in option on any OS (no install) → Vocal Reduction and Isolation effect. 1970s-style phase cancellation. Works on ~20-30% of songs. Honest answer: limited.

Audacity has TWO stem separation paths now (legacy phase cancellation + OpenVINO AI plugin), and they're not equally good. This guide walks through both, compares them honestly, and shows when external AI is still the right call.

Try StemSplit on your Audacity project →

Method 1: StemSplit (External AI, Best Quality, Every OS)

This is the cleanest, fastest option — and the only AI-quality option for Mac, Linux, and ChromeOS users (where the OpenVINO plugin is not officially supported). Browser-based, no install, drag-and-drop into Audacity.

Workflow

1. Upload your song to StemSplit → choose 4-stem (vocals, drums, bass, other) or 2-stem (vocals + instrumental).
2. Download the stems as WAV files.
3. Open Audacity → File → Import → Audio → select the stems. Each stem comes in on its own track.
4. Edit, mix, export — treat the stems like any other audio in Audacity.

The model is htdemucs FT — ~8.4 dB SDR in published benchmarks. Better than the OpenVINO plugin's stock Demucs v4 model on most modern productions, and dramatically better than the legacy Vocal Reduction effect.

Why this is the right answer for most Audacity users

• Cross-platform: works identically on Windows, macOS, Linux, ChromeOS, iOS, and Android. The OpenVINO plugin is officially Windows-only.
• No CPU/GPU constraints: processing happens server-side, so your 2014 ThinkPad runs the same model as a Threadripper.
• No install or build steps: you don't need to compile plugins, fight with Module preferences, or download multi-gigabyte model files.
• Cheap per song: a 3-minute track costs $0.30 in credits.

Common Audacity use cases

• Karaoke: extract instrumental → record voice on a new track in Audacity
• Vocal covers: get instrumental → record cover → mix in Audacity
• Podcast editing: pull vocal/dialog from interview audio caught with set music
• Restoration: separate stems → de-noise / de-click each stem individually in Audacity
• Sample creation: pull single-instrument loops from existing songs
• Education: import stems to teach mixing fundamentals in classrooms

---

Method 2: OpenVINO Music Separation Plugin (Free Local AI, Windows)

This is the most exciting addition to Audacity in years. As of Audacity 3.5 (May 2024), Intel ships a free OpenVINO-powered AI plugin pack that includes Music Separation alongside Noise Suppression, Whisper Transcription, MusicGen, and Audio Super Resolution. The Music Separation plugin uses Demucs v4 (htdemucs) — the same Hybrid Transformer Demucs architecture that StemSplit uses. Free. Local. Offline.

Cost

Free. Open source. Distributed officially through Audacity's website.

Install (Windows)

1. Update to Audacity 3.5 or newer at audacityteam.org
2. Download the AI plugin pack from audacityteam.org/download/openvino (a multi-gigabyte installer — model files are large)
3. Run the installer
4. Open Audacity → Edit → Preferences → Modules → set mod-openvino from "New" to Enabled → click OK and restart Audacity
5. The new effects appear in the Effect menu (e.g., Effect → OpenVINO Music Separation)

Usage

1. Import your song (File → Import → Audio)
2. Select the audio (Ctrl+A)
3. Effect → OpenVINO Music Separation
4. Choose:
   • Separation Mode: 2-Stem (Instrumental + Vocals) or 4-Stem (Drums, Bass, Vocals, Other)
   • OpenVINO Inference Device: CPU, GPU, or NPU (if your hardware supports it)
   • Shifts (Advanced): higher = better quality but linearly slower
5. Click Apply. First run takes 10–30s extra to compile the model for your device (cached after that).
6. New tracks appear with -Drums, -Bass, -Vocals, -Other suffixes

Honest comparison vs StemSplit

When the OpenVINO plugin is the right answer

• You're on Windows with a reasonably modern machine (Intel CPU with AVX2, or any discrete GPU)
• You need offline processing (sensitive material, no internet)
• You're separating a lot of songs and want zero per-track cost
• You're already in Audacity for the rest of the work

When it isn't

• You're on Mac, Linux, or ChromeOS: pre-compiled binaries are Windows-only. Linux requires building from source (non-trivial). macOS support is unofficial / experimental at best.
• Your machine is slow: CPU-only inference on older hardware can take 5+ minutes per song. Server-side AI is faster.
• You need the highest possible quality: StemSplit's htdemucs FT (fine-tuned) variant is slightly cleaner than the stock Demucs v4 the plugin ships with.
• You don't want to install a multi-gigabyte plugin pack and toggle modules in preferences.

The good news for free-DAW users: even if you're not on Windows or you don't want the install hassle, you still have an excellent option in StemSplit for $0.30 a song.

---

Method 3: Audacity's Built-in Vocal Reduction and Isolation Effect (Free, Limited)

This is what people usually mean when they search "remove vocals in Audacity". It's not AI — it's a phase-cancellation algorithm that's been around since the 1970s. It's free, built-in, and works on a small subset of songs. Worth understanding so you know when not to use it.

Steps

1. File → Import → Audio → select your song (must be stereo — mono won't work)
2. Select the entire track (Ctrl+A / Cmd+A)
3. Effect → Vocal Reduction and Isolation (in Audacity 3.x, sometimes nested under Effect → Noise Removal and Repair)
4. Configure:

5. Click Preview before applying — this is critical. If the preview sounds wrong, the technique won't work for this song.
6. Click Apply when satisfied → File → Export → Export Audio (WAV recommended).

Why results vary so dramatically

The Vocal Reduction effect assumes the vocal is panned dead-centre and is identical in both stereo channels. By inverting one channel and summing, anything centred cancels out. Modern vocal production almost never produces centred-identical vocals because of:

• Reverb (spreads vocal across the stereo field)
• Stereo doubling (a slightly pitch-shifted copy panned to one side)
• Stereo widening plug-ins
• Backing vocals panned slightly off-centre

When the vocal isn't centred-identical, the cancellation fails partially. And because kick drum and bass guitar are centred (always), they get cancelled too — leaving you with a vocal-leaky, bass-depleted mix.

Songs that work well in Vocal Reduction

If your song is from the era where this works, it's a legitimate free option. If not, use Method 1 or Method 2.

---

Method 4: Invert + Mix Technique (Rare — Requires Two Files)

If you have both the full mix and a matching official instrumental of the exact same master, you can perfectly subtract the instrumental from the full mix to leave the vocal. In practice, the matching instrumental almost never exists outside of internal label archives.

1. Import both files into Audacity on separate tracks
2. Zoom in and align them frame-perfect (look at waveforms)
3. Select the instrumental track → Effect → Invert
4. Tracks → Mix → Mix and Render
5. The matching content cancels; the vocal remains

When this works, it sounds incredible. When the master, length, or processing don't match (the common case), you get phasing artifacts instead of a clean vocal. Unless you have label-grade source files, skip this and use Method 1.

---

Method Comparison

For most Audacity users in 2026, the right pairing is:

• Windows + want free + okay with installing plugin pack → OpenVINO Music Separation
• Anything else → StemSplit

Both produce very good results. StemSplit gives slightly cleaner stems and works everywhere; OpenVINO is free at scale on Windows.

---

Pro Tips for Audacity Users

Use highest-quality source files. Both AI methods (and even Vocal Reduction) produce better results from WAV / FLAC / 320kbps MP3 than from low-bitrate YouTube rips.

Process at the song's native sample rate. Don't upsample to 96 kHz before separation — it wastes processing time without improving quality.

Keep stems on separate tracks. If you're going to mix or edit, leave the separated stems on individual tracks rather than mixing them down — gives you full control later.

Trim before separating. If you only need a chorus, trim first. Cheaper on StemSplit (per-minute pricing), faster locally.

OpenVINO plugin troubleshooting: If the plugins don't appear after install, the most common fix is Edit → Preferences → Modules → set mod-openvino from New to Enabled → restart Audacity. This trips up most first-time users.

Combine workflows. A common pro trick: separate with StemSplit (best quality), then use Audacity's Spectral Selection and de-noise tools on the resulting stems for final cleanup. Best of both worlds.

---

Frequently Asked Questions

Does Audacity have built-in AI stem separation in 2026?

Yes — through the free OpenVINO Music Separation plugin shipped by Intel and distributed via Audacity's official site. It uses Demucs v4 (htdemucs) and runs locally on CPU, GPU, or NPU. Important caveat: pre-compiled binaries are Windows-only. Mac and Linux users need to build from source or use an external AI tool like StemSplit.

What's the difference between Vocal Reduction and OpenVINO Music Separation?

Vocal Reduction is the legacy built-in effect — phase cancellation, no AI, free, works on stereo files where the vocal is centred. OpenVINO Music Separation is a free downloadable AI plugin (Audacity 3.5+) using Demucs v4. Completely different algorithms and dramatically different quality.

What's the best stem splitter for Audacity?

For most users, StemSplit is the best answer because it works on every OS, requires no install, and produces slightly cleaner stems (htdemucs FT). On Windows specifically, the OpenVINO Music Separation plugin is an excellent free local alternative.

Can I use Audacity's OpenVINO plugin on Mac?

Not officially. Pre-compiled binaries are Windows-only. macOS support is unofficial / experimental. Mac users should use StemSplit for best AI stem separation, or Logic Pro 11's built-in Stem Splitter if they have Logic.

Does the OpenVINO plugin work without internet?

Yes. The plugin runs entirely locally on your CPU/GPU/NPU. The only network needed is the initial plugin/model download.

Why does Audacity's built-in Vocal Reduction sound terrible on my song?

It's a phase-cancellation technique that assumes the vocal is centred and identical in both stereo channels. Modern vocal production uses reverb, doubling, and stereo widening, which all break this assumption. Use the OpenVINO plugin or StemSplit instead for AI-quality separation.

How long does the OpenVINO plugin take to process a song?

Depends heavily on hardware. With a modern Intel iGPU or discrete GPU: ~30 seconds for a 3-minute song. CPU-only on older hardware: 3–5+ minutes. First run takes 10–30 seconds extra to compile the model for your device (cached afterwards).

Can the OpenVINO plugin separate piano, guitar, and other instruments?

The current public release supports 2-stem (Instrumental + Vocals) and 4-stem (Drums, Bass, Vocals, Other Instruments). Intel has discussed an upcoming 6-stem model adding piano and guitar separation, plus drum sub-separation and dereverb — currently on the project's next branch.

Is StemSplit better than the OpenVINO plugin?

Slightly, on most material. StemSplit uses htdemucs FT (fine-tuned variant of Demucs v4) which produces marginally cleaner stems on dense modern productions. The OpenVINO plugin ships with stock Demucs v4. The bigger differences are deployment: StemSplit works everywhere; OpenVINO is Windows-only and requires a multi-gigabyte install.

Does StemSplit work on Linux and ChromeOS?

Yes. StemSplit is browser-based, so any device that runs a modern browser (Linux, ChromeOS, Mac, Windows, iPad, iPhone, Android) works the same. This is the main reason it's the recommended option for non-Windows Audacity users.

---

Get Production-Grade Stems for Your Audacity Project

Upload any track to StemSplit and import clean WAV stems straight into Audacity on Windows, Mac, Linux, or ChromeOS.

• Vocals, drums, bass, and other — as separate WAV files
• Works on every OS Audacity runs on (and on iPad too)
• Cleaner than Audacity's built-in Vocal Reduction effect
• Slightly cleaner than the OpenVINO plugin (htdemucs FT vs stock Demucs v4)
• No multi-gigabyte plugin install, no Module preferences toggle
• Free 30-second preview before paying

Separate Stems Now →