POST to speech endpoint fail on longer input values

[KingDanx]

From what I have observed if my input value is long enough (around 425 characters). I will get a 400 response and no audio is generated when stream is set to false.

If stream is not set to false I get a 200 response but no audio is generated

"steam": false

"stream":true

[remsky]

That's interesting, I've run it with entire book chapters without issue. Though I was running through the python requests library.

Will take a look and see if I can reproduce

[KingDanx]

More info on this:

The issue occurs when there is long bits of unpunctuated text, or long breaks between punctuation. The chunker.split_text() only seems to split text at punctuation. If there is no punctuation for a certain amount of characters, I think it should just yield a chunk to prevent the overflow error.

Examples:

No punctuation (no chunks):

Add 1 comma (Audio stops at the comma and overflows the rest)

Add 2 commas (Audio stops after the last comma and overflows the rest)

[remsky]

Should have the fix in the next merge. Looks like I didn't set up the fallback chunking quite right

[remsky]

Should be resolved now, if it happens again, please send in a new issue (I changed up the chunking quite a bit)