Wed Apr 30, 2025 7:30 am
I've just transcribed a doco with interview footage where someone is speaking with a reasonably strong accent. It coped with her voice fine for the most part, in that what it transcribed was generally correct, to a very useable level, but for some reason it would constantly miss the first word of sentences. Not get them wrong, just ignore them. So when the interview subject said
"So we grow fibroblasts for them, back in the lab"
I got something like:
"grow fibroblasts for them"
"in the lab"
Correctly timed, with a gap where the missing words should be. I mean it coped with words like "fibroblast", but couldn't pick up "So we" and "Back"
Also, is there a user control for the AI transcription that lets you choose which tracks to use for transcription? I had to mute the music and FX tracks before transcribing or my subtitles for the entire talkfest just said
[Music]
[Ship's horn]. (there's a ship's horn right at the end)
There is a music bed throughout, but it's mixed well down below the vox. I do use a ducker (built-in compressor with side-chain coming from the vox bus, that knocks about 10dB off the level) to ride the music levels, you might want to see if the transcription is getting mixed dynamic effects like that in its audio input or just the dry audio (I'd bet cash money that this is what's going on).
Resolve Studio 20.0b build 23, Linux
"So we grow fibroblasts for them, back in the lab"
I got something like:
"grow fibroblasts for them"
"in the lab"
Correctly timed, with a gap where the missing words should be. I mean it coped with words like "fibroblast", but couldn't pick up "So we" and "Back"
Also, is there a user control for the AI transcription that lets you choose which tracks to use for transcription? I had to mute the music and FX tracks before transcribing or my subtitles for the entire talkfest just said
[Music]
[Ship's horn]. (there's a ship's horn right at the end)
There is a music bed throughout, but it's mixed well down below the vox. I do use a ducker (built-in compressor with side-chain coming from the vox bus, that knocks about 10dB off the level) to ride the music levels, you might want to see if the transcription is getting mixed dynamic effects like that in its audio input or just the dry audio (I'd bet cash money that this is what's going on).
Resolve Studio 20.0b build 23, Linux