Give your ideas
a sound of their own.
Write a song. Shape an instrumental. Keep every version.
song, scene, or quiet moment.
Upload audio you own or have permission to use.
Whisper runs in your browser. First use downloads a recognition model and runtime (roughly 100 MB). Sung words, choir, and music may produce errors; review the draft before using it. No transcription API key or Terminal needed.
Reference guides sound texture, timbre, mixing, and overall feel. It does not guarantee the same melody, rhythm, or singer.
Experimental: repaints a padded ending, then joins its generated tail to your original decoded prefix with an 80 ms crossfade. The original file stays unchanged. If your snippet fades to silence, choose a keep-through point before the fade. Results may drift or have a noticeable join.
Source words plus your new section are sent as the full timeline lyrics. Placement and singer matching are approximate. Your existing new-song lyrics are not reused automatically. Choose Instrumental for a tail without new singing.
Total output = kept source + new seconds. Extension downloads as a re-encoded WAV. This does not transcribe your snippet automatically.
Creates music without sung lyrics. Your lyrics are kept if you switch back.
Generation options
Off by default. Cloud processing happens remotely; local mode uses additional memory. Vocal character comes from your sound description; no fixed singer or voice cloning is promised.
Paste your cloud key in Connections to begin.