Speech, music and effects splitter
Speech, music and sound effects as separate files. Speech comes out clean; music and effects split roughly.
- Runs on our server. Your upload is deleted within 24 hours.
- Space plays AB switch versions
You get four WAVs: speech, music, sound effects, and music with effects.
WAV · AIFF · FLAC · MP3 · M4A · OGG · up to 10 minutes and 500 MB
Checking your runs
Your file is processed on a cloud GPU by our processing partner and deleted there within 24 hours. We keep the results for 24 hours, keyed by the file's checksum, so a repeat in that time is free.
MP4 and MOV work too: only the sound is sent, the picture stays on your device.
How it works
- What it does
- Splits a soundtrack into speech, music and sound effects. Speech comes out clean; the music and effects split is rough, so you also get them together as one clean file.
- Model
- A stem model and a speech model on a cloud GPU find the voice; our server then splits your own audio with a mask, so every file keeps your stereo and the pair adds back up to the original exactly. Measured on 14 film clips with known parts (SDR): dialogue 13.8 dB, music and effects together 15.6 dB on average, where the untouched clip scores about −2 dB. Music and effects apart are a rough split (music 2.1 dB, effects 5.1 dB). Model names and licences
- Measured
- Measured on 14 film clips with known parts: speech 13.8 dB SDR on average. Music (2.1 dB) and sound effects (5.1 dB) are split roughly, against −3.4 and 3.2 dB for the untouched clip; for a clean bed, use Music and effects.
- Output
- Every file is a 24-bit WAV at your file's sample rate, and each pair adds back up to the original exactly.
- Your file
- Your file is processed on a cloud GPU by our processing partner and deleted there within 24 hours. We keep the results for 24 hours, keyed by the file's checksum, so a repeat in that time is free.
Speech, music and effects on their own
Clean speech
The speech file uses the same two-model check as the dialogue remover: 13.8 dB SDR on average on 14 film clips with known parts. The speech, music and effects splitter separates a soundtrack into speech, music and effects and also provides music and effects together. The speech is the stronger separation; the individual music and effects files are much rougher. Use those layers to understand or rebalance a scene, not as a promise of pristine studio soundtrack stems.
An honest music and effects split
Music measured 2.1 dB and effects 5.1 dB on the same clips: usable to hear what is where, not a clean stem. That is why the fourth file, music and effects together, is there. The combined background is an important practical option when you want to replace dialogue. Musical hits, drones and designed effects often share the same sound characteristics, making their individual assignment ambiguous. Keeping music and effects together avoids depending on that uncertain distinction while still reducing the spoken words.
Four files that add up
Speech plus music plus effects is your original, sample for sample, so nothing is lost when you rebalance. Complementary layers let you compare the split with the original and decide which balance serves the edit. Preserve timing and starting positions when importing the files. Rebuilding the input from the pieces is a consistency check, not proof that every sound belongs to the ideal category or that every isolated file is clean.
Who needs the three parts apart
Post-production
A rough M&E from a finished mix to start a re-edit or a dub when the stems are gone. A post-production editor can begin with the speech and combined background, then inspect the rougher music and effects only when a cue needs attention. This keeps the workflow focused on usable dialogue work. Obtain an original music-and-effects master when available rather than assuming separation can replace every production asset.
Podcasters
Pull a guest's voice away from a music bed and room sounds before you edit. A podcaster can examine an introduction containing speech, music and recorded ambience, then decide whether to keep the full bed or reduce it. Check soft words and transitions carefully. For a straightforward voice over music with no need for separate effects, the simpler background music remover may be easier to evaluate.
Researchers and students
Study how a scene is built: what the score does, what the effects do, where the lines sit. A researcher or student can hear how a scene places speech against score and effects. Write notes about uncertain assignments instead of presenting the output as a definitive sound-design breakdown. The model can place a rhythmic effect into music or a musical texture into effects, especially when both occupy the same passage.
How to split speech, music and effects
Drop the file
Audio or video, up to 10 minutes, 25 on Pro. Only the sound of a video is uploaded. Choose a short but representative scene containing the kinds of overlap you need to edit. Check the current duration limit before uploading a long sequence. A clean source matters: previous compression, clipping or aggressive denoising can remove details that help distinguish speech from the background.
Let three passes run
Stems, then speech, then music. A first run takes 2 to 5 minutes. Several processing passes are needed, so wait for every result to arrive before deciding that a file is missing. Compare the speech first, then the combined background, and finally the individual music and effects. This order makes the strong and weak parts of the separation clearer than auditioning an arbitrary file first.
Listen, then download
Every file plays on the same playhead; download each as a 24-bit WAV. Download the layers you need and preserve an untouched set before processing them further. For new dialogue, start with the combined music-and-effects file. If you use an individual effect, listen through its whole tail for musical leakage and check it again inside the destination edit rather than only in isolation.
Questions about the three-way split
Why is the effects file not clean?
No model we can run tells music from sound effects reliably yet; ours measured 2.1 dB for music and 5.1 dB for effects. We would rather say so than sell it as a clean stem. Listen to the combined background before deciding that the whole separation failed. It can be useful for dialogue replacement even when the isolated music and effect files are not clean enough to expose alone.
Which file should I use as a bed for new dialogue?
Music and effects together. It measured 15.6 dB SDR on average on 14 film clips, close to a real stem. Keep the rough individual layers for inspection when needed, but start the edit with the more reliable combined bed. Check the entrance and tail of every replacement line inside the full scene.
Is my file kept?
The audio is sent to cloud processing infrastructure, including Replicate, rather than being separated wholly on your device. Results may be held temporarily to serve downloads and identical cached requests. Save the output you need and consult the Privacy Policy for current retention and handling details; this result page should not be treated as long-term storage. Download the layers needed for your edit and keep the source recording separately. If the material must never leave your device, choose an offline process rather than this cloud separation workflow.
What should I check after using this speech, music and effects splitter workflow?
Listen to speech, music and effects and combined music-and-effects bed, rather than judging just one file. The difficult cases include the rough distinction between music and sound effects. Compare an exposed passage, a busy section and the tail of a sustained sound at the same musical moment. A result can work for practice while being too rough for an isolated production use. Choose based on the sound you will actually deliver, and keep the original nearby when a note, word or effect is uncertain.
What source recording should I use?
Start with the cleanest version you already have and avoid repeated lossy conversions. Saving a damaged recording as WAV does not restore the detail it lost. Excessive EQ, clipping or added reverb before separation can make the task harder. Keep the original and work on a copy so later comparisons remain meaningful. Treat category boundaries as editorial judgements. A sung word, a tonal alarm and a cinematic impact can occupy more than one conceptual category. Make a note of those moments and choose the file that supports your edit, rather than trying to force every sound into a perfectly pure speech, music or effects box.
How do file length, plan access and cloud runs work?
Targeted separation allows up to 10 minutes without a paid plan and 25 minutes on Pro; the chosen mode and uploader show the applicable limit. The current catalog allows 3 runs daily anonymously, 10 monthly for signed-in free users, 80 monthly on Pro and 150 on Pro+. Each new file and processing configuration uses the relevant allowance; an identical result can be free while cached. Check the current pricing and the uploader before processing a batch. Downloads are part of the result, not a separate per-file unlock.
Is this separation performed entirely on my device?
No. This is cloud separation: audio is uploaded for processing on external infrastructure. Browser playback, comparison and some exports happen locally after the results arrive, which does not make the separation itself offline. Check the Privacy Policy before submitting sensitive material, and use an appropriate local workflow if audio must stay entirely on your device. Download useful outputs while available; temporary results and cached repeats are not a permanent archive.
Can I publish or share the separated result?
Separation changes the balance of a recording, not its ownership or the permissions attached to it. Use material you own or are otherwise entitled to adapt, and arrange the permissions needed for your intended publication. A separated result does not include a licence to a commercial song, performance or soundtrack. For your own project, keep the source, an untouched separation and a separate delivery copy. Listen again in the complete mix before sharing; an acceptable rehearsal reference may reveal artefacts when exposed alone.