Audio still not consistently matching text in sentence mode

[Mycroft]

From the beginning, I’ve been in awe that the system does as good a job as it does of matching the audio to each sentence. But the beginning and or end of the sentence audio ends up short part of a word or long.

I just reloaded the audio in one of my chapters and regenerated timestamps, but it still comes up short. Of course, I’m not knowledgeable about how it works, but it seems to me that it should either be all over the place or spot-on. It’s not all over the place, so why can’t it match the audio exactly to the sentence?

I would like to advance to the next sentence and listen to it to see if I can understand it before looking at the text. This is frustrating when part of the word is missing, so I have no chance.

Alright, 1st world problems, and I LOVE the app, but still frustrating.

[Hsingh]

Did you try higher bitrate mp3s or M4a files.

The transcription of Ukrainian audio lessons is catastrophic

@BlueSkies305 If the MP3 is incorrectly encoded, our importer is treating part of the frame as silence. Firefox browser is less tolerant of that. Re-converting the file to MP3 with a different bitrate (256-320 should do the job) or converting to M4A should resolve the issue. You can convert with this online tool: https://online-audio-converter.com/

[Mycroft]

After you posted that, I tried converting the MP3 that I downloaded from Audible to M4A at 320 using Audacity. It was worse than before.

Then I downloaded the file from Audible again. Epubor gives the option of MP3 or MP4, so this time I tried MP4 and converted that to M4A at 320bps and it was still worse.

I don’t know what the bit rate would have to do with anything though. They’re still audio words that should match the text and if it’s clear to a human, it should be clear to AI. Starting the sentence audio early and ending it early should be fixable. Am I the only one with the problem? Maybe the AI just doesn’t like Harry Potter in Spanish?

[Hsingh]

Sounds like a new problem.

Previously it was the player in browser/app. They behave badly with low bitrate. I noticed the problem with timesyncs when playing the same sentence in app and browser. eventhough timestamp is same, the playback don’t match. Problem went away after using higher bitrate.

Will see what support says for this new sync issue.

[Mycroft]

I should have said that this was using the IOS app. I didn’t try to see if it worked properly in the browser. I’ll have to do that.

[Mycroft]

I tried it in the browser and the original audio sounds the same as on IOS. I already deleted the higher bit rate files and I don’t think it’s worth redoing all of that. The audio of the default MP3 from Audible should be adequate for the AI to determine the start and end of a sentence.