Bijan Bowen discovers unauthorized Mandarin versions of his videos on Bilibili and uses the problem as a practical test for Fish Audio S2.1 Pro. A 15-second sample produces a recognizable clone, while a longer 78-second sample improves stability for full-length generation.
He builds a local web workflow that slices a video, transcribes its speech, translates the transcript with a local Qwen model, sends Mandarin text to Fish Audio and places the generated speech back onto the original video. A short test proves the end-to-end pipeline works.
The first full-length render stretches short Mandarin lines across chapter-length windows and sounds unnaturally slow. Bijan changes the system to align smaller speech segments by onset and duration, producing a more natural dub and a reusable route for publishing his own translated videos. Sponsor thanks and promotional offers are omitted.
Watch on YouTube
