What WaveShift is
WaveShift translates and dubs finished videos. You give it a file or a link — YouTube, Bilibili, or a direct video URL — and it returns the same video speaking another language, with the original speakers' voices and the original background music still in place.
It supports English, Chinese, Japanese, Korean, Spanish, French, German, Portuguese, Russian, and Italian.
You can watch before it finishes
Most dubbing tools make you wait for the entire render. WaveShift streams the result as it is produced: once the first segment is mixed, playback can start while the rest keeps generating and joining on behind it. On a long video that is the difference between waiting for the whole runtime and starting in about the time it takes to make a coffee.
This is the single design decision the rest of the system is built around, and it is why the pipeline is segment-oriented rather than file-oriented from end to end.
How it actually works
Speech and background audio are separated first, so music and room tone survive the process instead of being flattened along with the dialogue. Speech is transcribed with speaker attribution, translated with surrounding context rather than line by line, then re-synthesised with a voice reference locked per speaker so one person does not change timbre halfway through.
Translated speech rarely lands at the same length as the original, so timing is solved explicitly: lines are fitted to their slots, and where a line cannot fit at a comfortable speaking rate it is rewritten shorter rather than rushed. The dubbed speech is then mixed back over the preserved background audio and packaged for streaming playback.
What WaveShift is not
It is not an AI avatar product. There is no synthetic presenter, no lip-sync retargeting, and no talking-head generation. If you need those, HeyGen and Synthesia do them well and we say so on our comparison pages.
It is also not an enterprise localisation suite with review workflows, seat management, and translation memory. It is a tool for getting a finished video into another language quickly and correctly.
You are billed for what is delivered
When a job starts, WaveShift holds an estimate based on the source video duration. When it ends, you are charged only for the dubbed audio actually delivered, and the remainder returns to your balance. The final charge never exceeds the estimate.
If a job fails for a reason on our side — a system, network, or unknown error — it costs nothing. Minutes are prepaid, never expire, and there is no subscription.
Who builds it
WaveShift is built and operated by a very small team, running on Cloudflare Workers with GPU inference on demand. That shape is a deliberate constraint: it keeps the service cheap enough to price per delivered minute instead of per seat per month.
It also means you are talking to the people who wrote the code. Bug reports get read by someone who can fix them.
Contact
Questions, bug reports, or something dubbed badly? Write to us — a real person reads it: support@waveshift.net
See also: What we shipped recently.