On Mon, Aug 31, 2026 at 09:49:40AM -0600, Sam Hartman wrote: > > Or how about if I move to using a llm-based TTS engine like (vibevoice > or qwen-tts) for my screen reader? > Do I really need to disclose that all my contributions are assisted by > AI because of accessibility technology I use? >
It might help: if the TTS gives you an incorrect output, that's worth knowing. If you say - my TTS gives me 94% correct output, please bear with me if I'm slightly slow as I work out the other 6% - that might be useful to others. > I'm moving towards having a transcription model (the one I use happens > to be an autoregressive transformer) take voice recorded blog posts and > transcribe them and then an LLM do some markdown formatting. > Especially if I review the final result of that and the LLM didn't > generate any significant text, I'm probably not going to disclose that > as AI generated. > If you use that, it's also worth knowing that you used AI. AI-mediated Markdown and then an output as WCAG 2.2-compliant HTML and CSS (which may be immaterial to you if you don't use it/because you can't see it). will mean that another partially-sighted person might not blame you if the output is unreadable in their screen reader and might not expect you to human-code to fix problems. Responsible disclosure is good and that's the result of the GR, surely. Anyway, debian-vote is not the place to still carry on discussing this in all likelihood. With every good wish, as ever, Andy > There are some cases where there are interesting boundaries. >
