Keep the words. Lose the soundtrack.
Separate the voice from a mixed soundtrack, listen before you export, and put the isolated voice back into your video. Best for spoken clips; music and speech artifacts can remain.
1 credit per started second, minimum 5. The video stays on this device; extracted audio is uploaded to process.
Your video preview appears here
Check quiet words, loud musical moments and the end of each sentence. Singing may remain; room sound may be removed.
Use a short video you own or have permission to edit. The picture stays on your device; only extracted audio is uploaded.
Check the measured duration and exact credit cost before starting voice separation. No prompt-writing required.
Listen for quiet words, breaths and dialogue under loud music. Environmental sounds can be removed too.
Build and download an MP4 with the isolated voice, or save the voice track for further editing. Keep your original.
These original scene illustrations show three practical uses. Audio comparisons use generated narration and an original synthesized music bed, processed with the same separation model. They are controlled examples, not a promise of results on every recording.
Read the case method, settings and limitations
Recover a spoken introduction when the instrumental bed competes with the product story. Check soft words and the end of each sentence.
Original mix
Isolated voice
Illustrated scene · synthetic narration · processed September 16, 2026
Make a short explanation easier to follow. Check numbers, consonants and instructions that overlap musical accents before publishing.
Original mix
Isolated voice
Illustrated scene · synthetic narration · processed September 16, 2026
Try extracting speech from a mixed recording. Room tone and other environmental sounds may disappear too; listen carefully before reusing the clip.
Original mix
Isolated voice
Illustrated scene · synthetic narration · processed September 16, 2026
Choose by the source you have and the deliverable you need. If your original editing project already has a separate music track, turning down that track preserves the untouched voice. A flattened export needs a different approach.
| Workflow | Best fit | Trade-off |
|---|---|---|
| Banana Pro AI | A short mixed video; guided voice isolation, comparison and MP4 export. | 30-second limit. Separation can affect ambience or speech, and singing may remain. |
| Timeline video editor | An original project with separate music and dialogue tracks; precise timing and manual mixing. | In a flattened export, a volume slider lowers music and speech together unless source separation is available. |
| Audio stem splitter | Music production or access to both vocal and instrumental stems. | For video delivery, you still need to place the selected audio back into the original video and check synchronization. |
Need to change the picture as well? Continue with the AI video editor. For a separate visual task, remove the video background or use the video upscaler.
Muting makes everything silent. Noise reduction usually targets hiss, hum or other noise. Source separation estimates vocal and instrumental components from a mixed soundtrack. It does not recover a pristine original voice recording, and it cannot reliably distinguish a background singer from a person speaking.
Listen through headphones at the same volume for both versions. Start with the hardest passage. If a name, number or consonant becomes unclear, keep the original or edit separate tracks. Heavy reverb, clipping and overlapping voices are challenging. If background ambience is essential, use a dialogue editor with manual control.
Source separation can reduce instrumental background music while retaining an estimated vocal track. Results depend on the mix: some music can remain and words can sound altered. It cannot reconstruct speech that was never captured clearly.
Not reliably. Voice isolation may remove these sounds too. If the ambience matters, use the original project with independent tracks or a dialogue editor with manual control.
The MP4 export copies your original H.264 video track and replaces its audio. It does not regenerate the scene. Upload an MP4 or MOV up to 30 seconds and 50 MB; your browser must support decoding its audio and encoding AAC. Browser encoding can add short padding or timing differences; check lip sync before publishing.
Singing and speech can both be classified as vocals, so the sung voice may remain. Preview the result instead of assuming all music will disappear.
Voice isolation costs one Banana Pro AI credit per started second, with a minimum of five credits and a maximum of thirty for a supported clip. The server measures your extracted audio and displays the quote before submission. Failed processing is refunded automatically.
Your account-scoped saved job resumes status updates without submitting another paid request. Keep the original video available and reselect that same file to build the final MP4. The isolated voice can be downloaded separately.
You can cancel local preparation before submission. Once the separation job is submitted, you can pause status updates but processing continues and cannot be canceled. A failed job is refunded; a retry after failure is a new paid job.
Processing does not change the rights in your source material. Only upload audio you are permitted to process. Removing music does not clear copyright or grant rights to reuse a recording.
Explore all AI tools or see credit plans before your next project.
