Week 94: Blank tape
My plan was to use the new OpenAI Whisper automatic speech recognition system so that I could record my blog post while walking, transfer it to my computer, convert it to text, and then probably edit it a bit because natural speech is usually more chaotic than the written word. I went out for a walk, and recorded everything I wanted to say. I got home and uploaded it to my computer, and found that I had recorded almost none of it, thanks to the audio permissions in Android. It captured a few seconds at the start and end, but in the middle where I’d switched to a different app before putting my phone away, it had recorded nothing but ten minutes of silence. It gave me a chance to rehearse what I wanted to say, perhaps. And it was a lovely day for a lunchtime stroll along the river. And it was so blustery that I was sceptical about how well it would come out, but even so, I was hoping for more than that.
I’ve been trying out Whisper so that I can transcribe the video of the talk I gave at LRUG last week. I felt at the time that it had gone really badly with all the technical problems. When I finally summoned the courage to watch it back, though, it hadn’t gone badly at all. Murray edited out all the gaps where I’d been waiting for a slide to update, or, in one case, waiting to get onto the internet again. With those bits taken out, although you can hear the joins if you listen carefully, it hangs together, it seems coherent, and I’m actually really pleased with it. It’s nice to know that despite my perception of the evening being a bit of a disaster, it actually went quite well.
Whisper works OK, and it’s tolerably fast on this computer. I followed these
instructions,
although I had to add a couple of flags: --fp16 False
to stop CUDA crashing,
and --model base.en
which is the largest model that fits in my 4GB graphics
card. I think the larger models probably handle less common words better.
I’m transcribing the talk so that I can turn it into a blog post, although I don’t know yet how I’m going to do the slides. I could export them to images, but what I’d really like is to have them as some kind of text so you can see the code easily. I haven’t worked out that part.
The 60p lettuce outlasted Liz Truss as prime minister. It feels as if the Conservative Party is running out of what meagre reserves of competence it ever had. Expelling everyone who wasn’t a rabid ideologue in 2019 didn’t help with that. It’s a wounded and intellectually enfeebled party, but it still manages to limp on instead of putting everyone out of its misery. For how much longer?
On Sunday, we went to a couple of artistic events in Southwark Park. The first was a choral and performance piece in Dilston Grove, O Complex Mass. I knew the venue was an old church (that much is obvious from the gothic windows and perpendicular proportions) but I didn’t know that it was “the first in-situ poured concrete building in the UK”. The piece was written for the space, and made excellent use of its distinctive acoustic properties.
At the start, it wasn’t clear that the performance had even begun. The singers were dressed in normal everyday clothing and held pieces of paper like the ones we’d been handed on the way in, and we couldn’t tell that they were about to perform. They started humming and speaking and making other crowd-like noises that coalesced into something gradually more recognisable until, after about fifteen minutes, they were gathered in two circles at different ends of the space, sometimes singing harmonies, sometimes making stranger noises. I loved it.
The second was a talk by Rosa-Johan Uddoh about the life and work of Una Marson, poet, playwright, and the first black producer at the BBC, by whom she was apparently treated very poorly. It was fascinating and moving.
We have half cleared the loft in preparation for improved insulation and boarding on Wednesday. The insulation seems like an even better plan with some of the more apocalyptic energy price forecasts, but it’s the increased storage space I’ll be happiest about. Unfortunately, that also means that the living room is crammed with boxes right now, and there’s yet more to come down before it all goes back again. It will be good when it’s done.