Emacspeak discussion list

[aleland.tech.list AT fastmail.com]

Hello Emacspeakers,

I recently upgraded from macOS Sonoma to Tahoe 26.2, and immediately noticed 
an unwelcome change in the way that SwiftMac 4.3.6 (Emacspeak 60.0; emacs 
30.2; using emacs-plus@30 via homebrew) handled longer sentences. All the 
voices I tried (Samantha, Tom, David; both enhanced and compressed) would 
suddenly pause at seemingly random places in a sentence, where there was no 
punctuation. This happened across emacs, in nov.el, org-mode, notmuch, 
anywhere I read text with longer sentences.

I isolated it by testing the same text three ways: in Emacs with my full 
config, in Emacs with emacs -Q loading only Emacspeak, and in TextEdit read 
by VoiceOver using the same voice. VoiceOver read smoothly; both Emacs 
configurations had the pauses. So it wasn't my config, and it wasn't the 
macOS voice engine.   

Then, using a method that I'm hesitant to admit may have included a look at 
the swiftmac source with the help of an AI coding tool (is this a faux pas to 
admit, let alone deploy? The etiquette around AI seems to be fraught at best 
in FLOSS-land), I discovered the chunkText function in main.swift, which as 
far as I can tell, splits all text into 15-word chunks, each sent as a 
separate AVSpeechUtterance. There's an inherent gap, my friendly tool 
informed me, between utterances, and Tahoe appears to have increased that gap 
enough to become audible, whereas it worked differently under Sonoma? 
VoiceOver may not chunk this way, which is why it's unaffected?

However, with the help of my confident, controversial, possibly problematic 
AI tool, I changed maxWords from 15 to 200 in chunkText() and rebuilt 
SwiftMac. This fixed the pauses for me! Wondering if anyone else had this 
prosody problem and if you're mad at me for using AI to troubleshoot and if 
this information is at all helpful.

As an aside, I had to abandon the Alex voice altogether, because after 
upgrading, both VoiceOver and SwiftMac have Alex pausing way too long after 
full stops so that problem must be a change in the TTS itself.

Off-topic epilogue: This experience of having my workflow disrupted so 
intensively (I read a ton of text in emacspeak for work and these random 
pauses really messed with my speed and comprehension) by a mere system 
upgrade has me considering trying out Linux as a daily driver, where e.g. 
Debian/Emacspeak/espeak-ng would seem to be a more stable workflow less prone 
to disruption by Cupertino caprice. Do you agree with this assessment?

Thanks as ever,
Andrew