epub2tts icon indicating copy to clipboard operation
epub2tts copied to clipboard

Clipping off or skipping the first word on a chapter

Open Lastofthefirst opened this issue 8 months ago • 2 comments

Kokoro has been such an improvement!

I am having an issue where the first word of a recording is clipped off, for example "In it's landmark message..." will be read as "its landmark message..."

This happens, not always, but fairly often. Any idea why or a solution?

Thanks, this has been so useful. I am "reading" so much more XD

Lastofthefirst avatar May 13 '25 17:05 Lastofthefirst

I've noticed that too, I think it depends on the voice.

Please give https://github.com/aedocw/epub2tts-kokoro a try if you have time. It might behave differently (that's the only one I've used in quite a while). Also try different voices and see if it does the same thing.

aedocw avatar May 14 '25 00:05 aedocw

I observe the issue with kokoro (am_michael and others) also depending on whether "skip-parts" feature flag is also set. I can add a few things:

  1. Depending on the voice, some only "clip" on consonants. Some clip on vowels.
  2. The before-and-after nearby phoneme elements can cause this
  3. I believe the solution is to adjust up by 0.5s the silence defaults at the edges, and provide the user the ability to adjust this silence, as it's a core narrative element, stylistically.

Either way, getting more isolation between waveform synthesis steps should solve this clipping issue (and several others.)

ken-stacktosea avatar Oct 20 '25 12:10 ken-stacktosea