Sometimes our narrators make errors (e.g. repeated words). Can we use aeneas to identify these so we can cut them?
Hi Karl,
Contact Andrew Shafe. (@sil.org). He has a group called SAB Assist that will go through entire builds and fine tune the audio. It may take a few weeks to couple of months, but the result is a report showing any anomalies.
We (Kalaam) try to send any build with audio to them for this service. But I warn you, it’s detailed and you may find more errors than expected. That can lead to pointing them out to the audio team for “retakes”; not something every team wants to hear. But it’s better to make it match as perfectly as we can.
Hi Karl (and others needing help with fine-tuning / error checking of audio files in SAB),
I don’t know of any way aeneas can check for errors in any meaningful way. As Dan mentioned, I’m the liaison with a volunteer team of checkers who will fine-tune an SAB project (even a whole Bible) through a crowd-sourced system. In the process, they check every syllable of every phrase and note any discrepancies in an Excel spreadsheet, fine-tuning the files as they go. As Dan said, the process can take a bit of time if there are many projects in the queue, but we can usually get an NT done in a couple of weeks once we get to it. If you’re interested, email me at andrew_shafe@sil.org. You’ll need to export the timing files and audio from SAB, but I can help you with that.
Thank you Andrew!
I was able to find --presets-word : apply presets for word-level alignment (MFCC masking)
Also these options look like they’re promising:
--faster-rate : print fragments with rate > task_adjust_boundary_rate_value
--zero : print fragments with zero duration
Please let me know if you have any feedback.