The latest episode is out and it is an interesting one.
Firstly, my guest is the first person to produce a laboratory confirmed intergeneric hybrid between raspberry and strawberry.
Secondly- they were too shy to speak in person, so I used an AI voice to stand in for them. I think you will agree it worked pretty well (and I am hopeful that this tool will help me interview all sorts of growers where internet access or english proficiency is a limiting factor). The AI did get strangely emotional at one point, but we both agreed to leave it in as a time capsule of what the technology could do in 2023. Enjoy!
The podcast is also now available on all major podcast distributors (for easier listening while you garden.
I’d say it’s not great Shane. Sorry. The voice is odd, lat emotionless, words are pronounced weirdly and the tempo of speach is off. Slow when fast would be ok and fast when a lot of complicated words follow up.
I’d have rather liked better if someone else or you read the answers out loud. Nothing wrong with keeping things low tech.
I’ll keep that in mind for future guests. I can probably get another person to stand in, or find a way to record the questions and the answers myself. Thanks for the feedback.
I listened to the whole thing. It was clearly not a human talking but I’m ok with that bc of the disclaimer at the start. Great idea for crossing language barriers with future guests.
I was producing an episode for a podcast, so audio format was necessary. Some podcasts have transcripts that go along with them, but that would add hours of work to every episode manually transcribing and correcting the text. Damien Beaumont has plenty of articles about his breeding projects on his blog linked in the episode description. You should check them out- there is loads more to the story.
I really enjoyed listening to the episode in this format. I find I learn differently with a podcast, and sometimes more (especially when walking). I found it hilarious when the voice became super enthusiastic about the dream plant breeding section.
What TTS AI did you use? Did you tell it to use an Australian accent? If you interviewed someone from say India, would you tell it to use an Indian English accent?
I used ElevenLabs, which I think is one of the better AI voices out there. If you spend a lot more time re-recording wonky sentences and editing it all together you can produce a really polished product. They had a new aussie male voice option this time so I had to give it a go- though it sounded a bit texan at times.
Personally, I honestly didn’t find the AI voice too off-putting or anything like some others apparently did. Yeah, it was obviously not a human, but once the episode got going, I got used to it enough that I didn’t notice it enough to disrupt my focus on or enjoyment of what was being said.
And unlike a blog post (which I do also see value in and will have to check out his blog sometime), I was able to turn it on while I was out in the back with a shovel and a pickaxe (to break up compaction!) trying to dig a hole…
After reading that it was a synthetic voice, I assumed it was going to be awful and that you were maybe too close to the project to evaluate it without bias. But honestly, I’m blown away by the quality. There are occasionally jarring mispronunciations like the hard G in germ and vegetable, but some of the others, I’m not even sure aren’t just differences between American and Australian dialect – like maybe the pronunciation of diploid (dee-ploid instead of dih-ploid at one point but not another). The fact that the synthetic voice pauses to rapidly inhale is a pretty keen touch. The pronunciation of what I guess is CSIRO, is a little all over the place.
It’s weird that as a child, he was told you couldn’t save tomato seeds! I’m continuously discouraged from growing e.g. apple from seed because they won’t come true, but no one tells me a plant won’t grow.
The berry hybrid story is so cool!
And yeah, the manic enthusiasm followed by short weird monotone toward the end is a fascinating artifact.
Yeah, after reading the comments and with some of my past experiences with AI text to speech, I was kind of bracing myself for the worst too - maybe that’s part of why I wasn’t bothered by it lol.
(At the same time though, while I was partly bracing for the worst, I have also seen how far some of the different AI technologies have come in the past couple of years and know that with a little bit of tweaking you can actually get some surprisingly good results. So I truly wasn’t sure what to expect.)
Yeah! I thought that was really fascinating too! Especially as a broad rule for the tomatoes that they were growing at home. I could see someone saying that now about a store bought tomato or other produce, but I was really surprised to hear that was the thinking about their tomatoes in general.
But it’s great that as a child he did what made sense to him and tried it anyway to see what would happen.
Thanks for giving the new technology a try. I agree real humans are still better, but the output from elevenlabs changed my mind on the potential for AI narration. I could have gone through and re-recorded all the weird bits by hand, but I think it is interesting to show people the flaws of the current technology as long as it is basically functional.
I found the episode to be very entertaining - ecstatic robot voice and all. We live in the golden age of plant breeding. Not only do we have access to germplasm from all over the world, we have the ability to translate languages in realtime. I have been reading books from the Soviet breeder Ivan Michurin - to think that if he was still alive we could both communicate. Keep pushing the boundaries.
