How to fix a generated voiceover that fades out at the end of its sentences
Tell Vizard Agent the words where the voice fades. It measures each generated line in short windows instead of trusting its average level, because a take can start loud and drop 10 dB or more by its last words. It evens out the level inside every take before placing it, then measures the finished mix line by line.
What do you type into Vizard Agent?
Quote the words where the voice drops away, as you hear them. That points Vizard Agent at one line it can measure. If you also want a different voice, say so in the same message, because a new voice means new takes, and the fix has to be applied to those too.
The voice fades at "a short walk to the beach". Can you make the voice more enthusiastic and softer?
What does Vizard Agent actually do?
It looks inside each line, not at the line's overall level. Generated narration often starts a sentence strong and lets it die away, and making whole lines equally loud leaves that slope exactly where it was. In the session behind this article it:
- Measured the raw take of the flagged line in half-second windows: about −19.5 dB at the start and −28 dB on the last words, so the fade was inside the generated take, not caused by the mix.
- Recast a brighter voice as asked, set every line to the same overall level, and measured again: one line still slid from −17 to −32 dB across the sentence.
- Ran speech-aware levelling on every take before placing it and printed each take's profile before and after: the worst line's slide of about 14 dB became a band of 4 to 6 dB.
- Measured every speech window of the finished mix, with the voice between −19.5 and −21.3 LUFS throughout.
- Transcribed the delivered file to confirm all 13 phrases were there, and had it reviewed for any line that still faded.
When does this not work well?
When the voice actually changes, not just its level. A take that turns breathy or whispery at the end still sounds tired once it is louder, so levelling fixes the volume but not the tone. Then regenerating those lines with a different voice or delivery is the fix.
Common questions
Is it my mix? Usually not. Vizard Agent measures the raw take first.
Why doesn't normalising fix it? It evens out whole lines. Vizard Agent levels inside each take.
Will it sound squashed? No. Vizard Agent levels gently and checks the result.