zalthor
Joined 15 years ago
Joined 15 years ago
One, in the post-transcription cleanup phase, instead of programmatically fixing the text, I just feed the text with a prompt + dictionary of words into an LLM. Initially tried doing this with the on-device model from Apple in macOS 27, but that was too slow. I've now been playing around with gpt-oss-20b on Groq, which appears to be fast and effective. Although I'm curious how well your programmatic approach works, I might try some sort of a bake off over the next few days. But in the long run, I think I might still stick with an LLM-based formatter because you can have far more interesting formatting considerations or formatting prompts based on the app that you are using or the style or tone of your dictated text.
Two - When building the dictionary of words to fix, I'm also including a recording of the word and having the transcription model just predict draft out variants to use as replacements.
What I find annoying from Sonic is their billing process. I dislike that they foist a random router onto you and charge you $10 a month (and charge you if you want to return it). I also dislike that they force a landline on you and charge you another $10 on you. Who in 2019 needs a landline? What I find somewhat dishonest, is how the sell it. I get that most advertising is not supposed to reflect reality, but it's sort of strange that you get excited about ads that say and "1GBPS internet at $40 a month" and only a year later realize that you're paying $75 a month instead. And that's something that's public, who knows what else about their business is as deceptive.
I suppose that's why Sonic has survived. I suppose that's the kind of business you need to run if you want to have a successful ISP.
Also, you can get it to connect to your phone via a voice command ("Alexa connect <device name>"), when you want to use it.