Free Vocal Remover: Get the Track You Need

A free vocal remover is useful only if you leave with the right file. To remove vocals from a song for singing practice, download the Instrumental track. To isolate vocals for editing, download Vocals. UVR Online produces both in a standard separation; you do not need a second job to get the other part.

This guide follows that whole-file workflow and reports three real processing runs, including complete downloads.

“Remove the lyrics” usually means removing the sung voice. Text burned into a video is a different problem: audio separation does not erase on-screen words.

On the result page, the original and generated tracks share a timeline. If you can still hear the full singer, first check which tracks are audible: the Original track contains the original voice by definition.

To inspect the backing track, mute Original and Vocals and leave Instrumental audible. To inspect the acapella, mute Original and Instrumental. Use each track’s play, mute and volume controls deliberately; several audible tracks can overlap. Move within the waveform or loop a short passage to compare the same moment instead of repeatedly guessing where it was.

Download the chosen stem with its own download button. A browser volume or mute setting is a monitoring choice; it does not automatically change the mix in the downloaded file.

UVR Online is an independently run online version inspired by Ultimate Vocal Remover. We offer a free allowance for light, everyday use. If you prefer to run the software on your own computer, you can also use the open-source Ultimate Vocal Remover GUI v5.6.

GitHub project · Prebuilt installers and executables

Not sure which to use? Read our UVR Online vs. the Ultimate Vocal Remover desktop app.

Free input: MP3 or WAV; credit-funded input also supports FLAC and M4A. Free output: MP3 or 16-bit WAV; credit-funded output also includes FLAC and M4A. Select one export format before processing. Downloading a generated file again is free; other formats are not created automatically.

Free access has file-length, file-size, frequency and abuse-prevention limits. Check the limits shown in the uploader before choosing a file: they can be adjusted by the operator, and network-level limits can apply as well as browser or account limits. “Free” does not mean unlimited automated processing.

For longer files, frequent work or batch queues, check current plans and limits. Paid formats also include FLAC and M4A. Choose the plan for the work you need rather than assuming a larger package will make the same model separate a song better.

1. Free processing accepts MP3 or WAV. After upload, choose any online model and MP3 or 16-bit WAV output, then click Start separation. No registration is required while your allowance remains.
2. After the full result is ready, compare the difficult passage with the original. Check both the wanted detail and unwanted leakage before choosing a track.
3. Free input: MP3 or WAV; credit-funded input also supports FLAC and M4A. Free output: MP3 or 16-bit WAV; credit-funded output also includes FLAC and M4A. Select one export format before processing. Downloading a generated file again is free; other formats are not created automatically.

If your goal is a sing-along track, the karaoke workflow covers that next step. For voice-only editing, use the acapella guide instead of running the same upload again.

On 6 November 2026 (Beijing time), we processed three files through the production service using an operator-created internal free-tier account and the ordinary queue. The two MP3s were the first three and five minutes of the same licensed song, West Wind. A third test decoded the first minute of that MP3 into WAV. These are three file tests of one song, not three different musical styles.

All runs used standard mode, kim-melbandroformer, an NVIDIA GeForce RTX 4060 (8,187 MiB detected), PyTorch 2.5.1+cu121 and CUDA 12.1. Segment length was 4 seconds and overlap 0.25. The worker was already running; we did not measure a controlled cold start.

How to read the timings: upload is the test client’s direct transfer to object storage. Worker processing includes separation, output encoding and waveform generation; it excludes network transfers. Total time starts when the upload request is initiated and ends when completion is observed. It includes validation, queueing, the scripted processing confirmation and transfers, but excludes downloading the finished files. Status was polled approximately every three seconds. This was an API-driven run through the real task workflow, not a timed browser click-through.

For the five-minute MP3, the two complete outputs were 300 seconds each. Both were actually downloaded: 8.3 seconds for Vocals and 9.2 seconds for Instrumental over this test connection. All six stem downloads across the three jobs passed file inspection and byte-range checks. The WAV outputs were confirmed as 16-bit PCM.

The internal account used no paid priority and no payment was made. This test did not exercise anonymous CAPTCHA, email registration or every free-access restriction. These are single-run observations, not an average, a maximum wait or a speed promise. Other users’ queue activity and a different network can change the total. Without clean reference stems, these tests cannot establish an accuracy percentage or superiority over another service.

The one-minute WAV test is useful precisely because it was not the fastest end-to-end run. A shorter song does not guarantee a shorter wait when file transfers and queueing contribute to the job. Its two WAV outputs also use much more space per minute than the MP3 outputs.

At 44.1 kHz, 16-bit stereo PCM, five minutes contain about 52.92 MB of audio per WAV stem, or 105.84 MB for two, before small headers. That is a format calculation, not an additional five-minute WAV test. An MP3 input can be small because it is compressed; separating it does not preserve that compressed size when the output is PCM.

Choose MP3 when transfer size and an everyday practice copy matter most. Choose WAV when your next editing step needs PCM audio. WAV avoids another lossy output encoding step, but cannot undo input compression or remove separation artifacts. See MP3 versus WAV for stem separation for the format trade-offs. File sizes here use decimal MB, not MiB.

Separation estimates sources from an existing mix. A trace of singing in the instrumental is vocal bleed; a softened guitar or a clipped consonant is a different trade-off. Neither is fixed simply by renaming the file or changing its extension.

Check that you downloaded the intended stem, then check the same passage in the original. If the problem is genuine, use the vocal-removal troubleshooting guide before spending time on repeated runs. If you need drums or bass individually, standard two-stem output groups those instruments together: read two-stem versus multitrack separation to choose the appropriate operation.

Can I remove vocals from a song for free and keep the entire result?

Eligible free jobs on UVR Online return complete results within the current allowance. Check the uploader’s limits, select MP3 or 16-bit WAV, and download before the displayed deletion time. More frequent or longer processing may require paid capacity.

Is a vocal extractor the same as an instrumental maker?

They describe opposite sides of this separation. An acapella extractor keeps Vocals; an instrumental maker keeps Instrumental. To split a song into vocals and music, save both outputs from one run.

Can I remove background music and keep a speaking voice?

A song-separation model is not a guarantee of clean dialogue recovery. For spoken recordings, see keeping voice while removing background music before choosing a workflow. Do not interpret successful singing separation as a speech benchmark.

Can I upload a music-video link or remove lyrics from a video?

This audio workflow accepts local audio files, not arbitrary remote URLs or video links. If you have permission, export the audio in a supported format first. Removing a sung voice will not remove captions or video graphics. We plan to add file imports from remote URLs in the future. This option is not available yet; for now, please upload a local audio file.

Start with the free vocal remover. Use only material you are entitled to process; making an instrumental does not grant permission to redistribute the song.