One song. Two tracks.
Isolate the vocals. Keep the instrumental. Make something new.
Your audio
Your audio is sent to Hugging Face for separation. Results are available here for a limited time.
Separated tracks
WAV outputChoose a song to create your two tracks.
Vocals
The voice, on its own.
Instrumental
Everything behind the voice.
Compare each result with the original using the players. Separation quality varies by song.
Speech into text.
Whisper large-v3 listens to your audio and writes down every word.
Your audio
Your audio is sent to Hugging Face for transcription. Language is detected automatically.
Transcript
Plain textChoose a recording to read it back as text.
Any words. Any voice.
Write the line. Add a reference clip and the voice comes with it.
Your script
0 characters ·
Advanced settings
Your text and any reference clip are sent to Hugging Face. Only use voices you have the right to clone.
Generated speech
WAV outputWrite a line, then generate it.
Speech
Your script, spoken.
Cloning quality depends on the reference clip: clean, single-speaker audio works best.
Silent video. Real sound.
Describe what it should sound like, and the clip comes back scored.
Your clip
Advanced settings
Your clip and prompt are sent to Hugging Face. Generation is capped at 30 seconds of audio.
Result
Video outputAdd a clip and describe the sound you want.
Pull it apart.
Dialog, effects and music on separate tracks — or every speaker on their own.
Your mixture
Send video and each track comes back as video carrying only that sound. Speaker separation returns however many voices it finds, up to four.
Tracks
Up to 3 tracksChoose a mixture to pull apart.