This blog post is a test for something new I’m working on. I think a lot of people feel overwhelmed that AI is everywhere right now. Personal blogs should lean in to the human voice. There’s a new play button on the permalink for this post on the web which will use a recording that I’ve uploaded.
@manton how many people are going to say that’s really good training data for a voice model…. But it was nice to hear your personal voice, as another human.
@phils Thanks! Yep, it's a great question. Voice models are about to get very, very good. But they're still not me.
@manton How do we know you recorded that audio and are not using AI that used your voice to train it? 🙃 With your podcasts, I bet it’s pretty easy to do and quite accurately.
@manton I don’t see any play button (Chrome-like, desktop). No sign in the source either…?
But - https://podnews.net/update/podcast-show-final has a “LISTEN” button just below the main graphic - and look what happens when you press play… you get a transcript! Surprisingly easy to do.
@pratik Heh. Well, the mistakes I didn't edit out this time. 🙂 But yeah, synthetic voices are getting incredibly good.
it's over here with all of the comments
@james Oops, I see what happened. The “transcript” link messed up the post on Mastodon. Fixed going forward.
@manton But jokes aside, if I train an AI model with my voice and use it for such post narration, is that not using AI "correctly"? It will save me tons of time and yet personalize my blog posts. Heck, I don't mind (in fact, prefer) if it preserves my stutter.
@pratik Maybe it’s like alt text: hand-written alt text is great, but AI-generated text is so good why not use it so that it’s accessible to more people? Likewise maybe in the future all posts will have synthetic narration by default but humans can override as needed to provide their own audio.
@numericcitizen @manton cool feature. I’ll update the Cards Theme with support for it when it’s released.
@manton Yup. Auto-generated (in my voice) but the option to override specific posts in case I want to do it in a different tone/mood/impression.
@manton This is cool work. Out of curiosity, in your example post, is AI transcribing your voice input? To me it sounded very much like you were reading the post as if it were already written. If it was transcribed, how did you handle the screenshot image insertion? Was it the little “screenshot here” aside? Finally, does the transcription save as a draft to clean up then manually post? Sorry for many questions.
@jimmitchell I wrote the blog post myself, then read it, but funnily enough after I posted it Micro.blog did transcribe the audio back to text with AI. In this case, that was redundant, so the transcript can be tossed.
@manton Interesting. I can't seem to find any transcriptions for my tests on my test micro.blog. Are you dumping them as part of the feature, or have I missed something? Will the feature ever get to the place of being able to upload audio and the post gets transcribed?
@jimmitchell One limitation that you might be hitting: we only transcribe one audio file per day. I was worried about the costs, but it has been fine so I'm going to raise that limit this week. If you click Transcripts, you can delete previous transcripts to work around the limitation for now. Also, check Account → View logs. It has log entries specifically for transcripts.
@pratik Yes, I actually do this pretty often! Upload audio to my test blog, grab the transcript, then delete it.
@jimmitchell Also check out the other limitations in the documentation. For example, it only works with MP3 files.
@manton What? RTFM? Never! 🤣
Thanks. That clears things up a lot and explains why I didn't see a transcript, since I uploaded in .m4a format.
