Like lots of other folk I’ve been reading plenty about Large Language Models, AI & Chatbots and playing with some of the toys.
I really liked Professor Bender’s approach and method. I also found this a very easy listen. My mind has tended to wander off when reading blogs post about AI. Very clear on the “not intelligent” and the risks associated with chatbots trained on large piles of language.
And specifically the things that they’re predicting is what would be a plausible next word given all the preceding words here and then again and then again and again.
And so that’s linguistically interesting that once you get to billions of words of text, there’s enough information in there just in the distribution of words to stick with things that are both grammatical and seemingly coherent.
So that’s a cool observation and it’s dangerous because we tend to react to grammatical, fluent, coherent, seeming text as authoritative and reliable and valuable.
So instead of talking about automatic speech recognition, I prefer to talk about automatic transcription because that describes what we’re using it for and doesn’t attribute any cognition to the system that is doing the task for us.2
I subscribe to the RSS feed of this BBC radio program as a podcast, pity you can’t find the feed on the webpage. ↩
Ironically I used Aiko to get the text of the podcast for the quotes: “transcription is powered by OpenAI’s Whisper model running locally on your device” ↩
The above is a short selection of a few moments of the stream I broadcast this morning, 10 minutes in total. The segments are each separated by 1 seconds of generated silence. The Whole thing is on the Stream Page. There is a fox barking near the start and a pigeon claps its wings over head near the end.
First, an apology it seems I should have streamed on Saturday. I somehow got mixed up and though it was Sunday. So 4am saw me leave the house and head for the park. Set up and started streaming at about 4:20am.
The birds were singing and there was not much traffic. As well as my phone I’d brought along my wife’s, to listen to the main stream and watch the chat. The Shure MV88 mic I use does not have a place to monitor the audio or listen to other audio on the phone. The borrowed phone worked ok, until listening for over an hour burned through my wife’s data 🙁
Not having noticed my scheduling error, I wondered if I was an hour out and asked in the chat about that. Hannah graciously didn’t mention I was out by a day and asked if I could hold on till 5:45. That I could. I’d could see I was live, but had trouble listening, but read in the chat it sounded ok.
I had listened now and again the day before and to some of the long running streams. I listened a lot more during the hour or so I was in the park. It was quite strange listening to the local sounds in one ear and England, the Netherlands, France & Barcelona in the other. Next time (and I’d love to do this again) I’d listen more during the whole 24 hours and pay more attention to the chat.
There is a well developed community around soundcamp.
I suppose I’ve done a fair amount of very casual field recordings over the years. Starting with audioboo, through SLFTalk, #ds106radio and Edutalk. Some of this has been from a computer at conferences & TeachMeets mostly from a computer. The mobile has been pretty casual andI’ve not worried about monitoring anything. Taking my wife’s phone was a lost minute idea, it might be an idea to join more dots.
I’ve used a variety of iOS app for streaming in the past but soundcamp have their own, locuscast. They had some last minute problems with a server which meant a switch to iziCast. Icecast clients for phones come and go and are of variable quality. iziCast seems more professional than many others, not free, I hope it stays around for a while. It can be used as a recorder as well as a icecast streamer.
Some of the other options look inviting, a raspberry pi perhaps. I’ve also got a zoom H4 N mic somewhere which would hook up to my phone and allow monitoring. I should wait till my old lightning port iPhone is replaced in case USB C comes to iPhones before I buy a connector.
One of the most interesting things about this was being out so early. We enjoyed walking the park before 8 during lockdown. I usually like to get out early on hillwalks, but this was much earlier. I saw a couple around 5:30 walking their dog and someone was feeding herons about 6 but other than that things were very quiet.
I also spent about an hour and a half in a couple of square metres. Usually if I am outside I am walking. I need to think about sitting still more often. Also worth going further afield. Also worth working a bit on learning some more bird calls. I only recognise a few.
I mostly forget that this blog is supposed to be about teaching & technology. There might be a few clues and ideas that could be imported into the classroom. I’ve alway liked online collaboration between classes. Could do something similar. Bird sounds or even playground sounds across a range of schools might be an interesting broadcast.
As soon as I stopped streaming a few more birds appeared. A couple of coal tits and a goldcrest were flitting around the bushes behind me. To early for a clear photos but nice to see.