#23: The Translated Episode

In the last episode before the summer hiatus, Ian and Michael discuss translation and speech in the context of language learning, virtual voice-overs, and virtual movie actors. As an experiment, we’ve also included excerpts from episode 16 (No code and low code) that have been automatically translated to Spanish and French using speech-to-text (to create the transcript from the original English), machine translation, and text-to-speech!

AI Unfiltered will be back from its summer hiatus with new episodes on August 30!

Ian Bowie
Hello and welcome to AI unfiltered with me, Ian Bowie, and our resident expert, Michael Stormbom. Where we will be talking about everything to do with AI in our modern digital society and what the future holds for all of us.

Michael Stormbom
I don’t know if we’ve talked about, I think we have actually, the possibility of automatically translating for example this episode into other languages for people who don’t speak English why not? Yeah. We should make an experiment because of course, we have the we have the transcripts, of course using speech recognition technology produced. So we could take that, and put it into machine translation. Then we put in our speech synthesizer and then we see what comes up and then we can.

Ian Bowie
Would be quite funny.

Michael Stormbom
And we can of course also create speech synthesizers that mimic our voices.

Ian Bowie
Wow.

Michael Stormbom
Because of course, while we’re recording these, these episodes, of course, we have tons of audio material on ourselves.

Ian Bowie
This is true. Yeah, absolutely.

Michael Stormbom
There was that, now I forgot the name but there was someone who had created a speech synthesizer and he’s taken like different TV stars. You could, for example, Patrick Stewart, so you could feed in a text and you could.. So Patrick Stewart in in the role of Picard. Of Star Trek Next Generation, so you could you could get him reading up, well, basically whatever you want.

Ian Bowie
Wouldn’t there be some kind of like, Patrick Stewart popping up saying you can’t use my voice like that?

Michael Stormbom
I reckon Patrick Stewart could probably have something to say about this. But yeah, to my knowledge, he hasn’t said anything. Right. Okay. Actually, we we spoke about Anthony Bourdain the other day, and actually, you know, there was a documentary about him, released some time after his death. And they actually have sort of like a fake Anthony Bourdain. So using speech synthesis, reading up some material, so he’s sort of like he’s talking in the documentary, but it’s not actually him because It’s just a speech synthesizer that have been trained to to sound like him.

Ian Bowie
Yeah, I read about that. But didn’t they also have to get permission from the family to do that?

Michael Stormbom
I think so. Yes. Yeah.

Ian Bowie
I think so. Yeah. So you can’t just go and do it.

Michael Stormbom
No, no, you cannot. No. Yeah. I mean, there’s also brings to mind these, you know, the good old Star Wars movies series. There was one of them. So they, so one of the actors that they wanted to use, or one of the characters that they wanted to use, so the actor had died, and well then they had a computer generated image of him. So it’s like, well they didn’t use speech synthesis, but they used an impressionist to make it sound like him. It wasn’t appearing, you know, entirely natural looking, you could tell it was a a CGI character. I think that ruined a little bit the suspension of disbelief, but.

Ian Bowie
It’s amazing what technology can do, isn’t it?

Michael Stormbom
Yeah, no, for sure.

Ian Bowie
But of course, the danger of that in the future is you know, they talk about fake news now. I mean, we’re gonna get super fake news in the future, aren’t we? I mean..

Michael Stormbom
Yeah, if it becomes so good that it’s indistinguishable from reality. Yeah. Then we’re gonna have a lot of trouble.

Ian Bowie
Can you imagine you know, like, setting up a fake Joe Biden?

Michael Stormbom
Yeah. So I mean, there was, you know, when the deep fakes entered the mainstream consciousness, I mean, they made that fake of Barack Obama. Yes. Yeah. So imagine that they were so good that they’re entirely indistinguishable from.

Ian Bowie
Yeah, scary. Yes. So they can do anything?

Michael Stormbom
Yeah, get Joe Biden go, drop the nukes on Turku right away. We have done nothing to deserve this. We’re perfectly nice people.

Ian Bowie
Yeah. But actually talking of translations, this sort of camera translation. That’s interesting to me.

Michael Stormbom
Yeah, so I mean, that’s been around for quite a while. Yeah. So basically, you can think of…

Ian Bowie
I’ve been asleep.

Michael Stormbom
Your camera phone can take a picture and can translate the text and, or it can even do it while you’re holding up the phone, and then it just translates the text.

Ian Bowie
It does it in real time now? The Google one?

Michael Stormbom
Yeah. So I mean, that’s handy if you’re in a foreign country, you don’t speak the language. You can just use the phone and…

Ian Bowie
Stick it over the restaurant menu. For example, for example. Yeah. Yeah. I think that’s actually really quite cool.

Michael Stormbom
Yeah, very much. So. That just goes to show how far the technology has come.

Ian Bowie
But somebody was telling me also that, they think, they’re not entirely sure that there is there is another one of these softwares, which is camera based. And you actually point it at an object. And it tells you what it is.

Michael Stormbom
Yeah. I mean, that’s a… well, I mean, we’ve talked about computer vision on the podcast before so yeah, I mean, it’s perfectly possible to do with some degree of accuracy, yeah for sure. Yeah.

Ian Bowie
So you know, the idea was, you point it at a carton of milk. And it tells you…

Michael Stormbom
it’s a carton of milk. Yeah. Yeah.

Ian Bowie
I mean, that’s actually quite funky. It’s also more immersive, because I mean, you know, if you think about these programs or apps like Duolingo, you know, that is not interactive at all.

Michael Stormbom
Like from a language learning perspective, you could have a an app in your phone that, you take a picture you show, show a camera and it tells you what the name of the object is in any language.

Ian Bowie
Yeah. Brilliant. But no, I mean, the you know, the technologies to assist people with languages today are just incredible. And they’re only going to get better.

Michael Stormbom
Yeah, there’s so many things you can do, do with it.

Ian Bowie
And then this speech. Of course you work with speech to text anyway, but the speech translators that you know, you talk into it in Finnish and press a button and translate it straight into English?

Michael Stormbom
Yeah, I mean, well, once it works at a reasonable level. Yeah, I think to some extent, it already does. I mean, they will change communication quite…

Ian Bowie
Absolutely. Absolutely. Yeah. Yeah.

Michael Stormbom
Yeah, so as I said, when we’re doing the podcast, because we can do it simultaneously in multiple languages, possibly.

Ian Bowie
Brilliant. Be a first I think, probably, wouldn’t it? Yeah. So have to be the first to do that.

Michael Stormbom
I don’t know. But let’s for the sake of argument say…

Ian Bowie
A first on a podcast.

Michael Stormbom
Yes. It’s the world first on this podcast.

Ian Bowie
You heard it here first. Wow. Oh, yeah, that will be… That’s just it, isn’t it. Ro do that, just in terms of working hours. How long would it take?

Michael Stormbom
Well, I mean, if we have the transcript, that takes what? Well, I mean, of course, it’s speech technology, that’s nearly instantaneous. And then of course, if you want to do some corrections, before manual, let’s say half an hour for manual corrections, and then we transfer into the machine translation. That’s instantaneous, basically. And then, depending on how much corrections we can do with… But let’s for the sake of argument say we do it entirely without any human interaction at all. Yeah, any no human like editing whatsoever in each of these steps. So we do speech to text, no human editing. Machine translation, no human editing, and then speech synthesis. So then it’s in principle. Well, almost instantaneous.

Ian Bowie
Blimey. Wow. That’s so cool.

Michael Stormbom
Yeah, well, let’s see how good the results are..

Ian Bowie
The whole idea that you can even think that you might be able to do it is cool.

Michael Stormbom
And of course, in the future, there might even be this sort of, like, setup where it translates directly from, like speech to speech. So there’s no intermediate step even necessarily. Yeah. Because of course, I mean, you can do that as well. It’s called an end to end system where you just…. you don’t necessarily need to have this intermediate step with machine translation. Like if you’re if you have audio and an audio of what it is supposed to be like, you can train it to do that as well. Just why not? I have no idea how much data you would need to do it, but in theory could be done. Yeah.

Ian Bowie
So actually, I mean, you know, not not only does this podcast become just about, you know, AI and all things technology driven, but we also become a language learning platform as well.

Michael Stormbom
Yeah, assuming we get it to work, but in principle, yes.

Ian Bowie
Well, that’s how all great things start, isn’t it in principle?

Michael Stormbom
Yes, yes, we dare to dream and make it reality.

Ian Bowie
So what we’ve decided to do, is we’re going to take a couple of clips from some of our episodes, and we’re going to use machine translation to put them into French and Spanish

Michael Stormbom
That is the plan. And so the concept here is that first that we do speech to text, so we take the transcript in English, we’re using speech recognition, but then we don’t do any editing to the text or just whatever the speech recognition puts out. That’s what we’re gonna put into the machine translation. And then likewise, when you put into machine translation, so we translate the script. No, no editing whatsoever. And then we put that to speech synthesis to produce the audio. And then we see what comes out on the other end, so to speak.

Ian Bowie
So if there any naughty words in Spanish or French, it has nothing to do with us-

Michael Stormbom
We take no responsibility for it. It’s the fault of the machine. Yes. Before we put it into episode, we will just take a look, we’ll just do a little sanity check to make sure that they can actually include it. But otherwise, if it is complete nonsense, then it’s complete nonsense, but.

Ian Bowie
And if it works, we’re going to experiment with all kinds of other languages as well.

Michael Stormbom
Yeah, I think I mean, because we do have a reasonable working knowledge of Spanish and French so we can sort of verify the content. But if it goes beyond that, then I think we… maybe we can crowdsource that. It would be interesting, but we would have no way of assessing the quality.

Ian Bowie
We might get some pretty quick feedback.

Michael Stormbom
You might get some angry feedback.

Michael Stormbom
Okay, so here comes the translated part of the episode. So we have taken the first four minutes of episode 16 on no code and low code and translated that into Spanish and then into French.

Michael Stormbom
And just to repeat the disclaimer, these translations have been done entirely automatically. So we have not corrected any translation errors or anything of the sort. This is done purely as an experiment. But without further ado, here comes to translated part.

Spanish Virtual Ian Bowie
Hola y bienvenidos a AI unfiltered conmigo, Ian Bowie, y nuestro experto residente, Michael Stormbom. Donde hablaremos de todo lo que tiene que ver con la IA en nuestra sociedad digital moderna y lo que nos depara el futuro.

Spanish Virtual Ian Bowie
He pensado que hoy sería muy interesante hablar de algo que acabo de leer, que es la IA de bajo código o sin código, y el hecho de que entonces se puede conseguir que personas no técnicas creen algoritmos basados en la IA sin saber cómo se construyen realmente.

Spanish Virtual Michael Stormbom
Sí, en efecto, y no se aplica sólo a la IA, sino que la regla general de no código, de bajo código, de movimiento, es que no se necesita necesariamente mucha habilidad de programación o, en el escenario ideal, ninguna habilidad de programación en absoluto para producir, bueno, software o aplicaciones o, de hecho, IA. Así que…

Spanish Virtual Ian Bowie
Sí, es decir, creo que eso me parece increíblemente estimulante. Si pudieras escribir, o incluso grabar, o una combinación de ambas cosas, los parámetros que quieres establecer, el propio programa haría la magia.

Spanish Virtual Michael Stormbom
Sí, absolutamente. Y, por supuesto, quiero decir que una gran parte de la IA es, en cierto modo, unir las piezas de Lego. Hay procesos muy estandarizados, así que incluso, incluso la creación de la IA como algo que podría concebirse, quiero decir, hasta cierto punto ya está automatizada también.

Spanish Virtual Ian Bowie
Sí, es decir, sólo por interés, ¿a qué distancia crees que estamos de eso como realidad?

Spanish Virtual Michael Stormbom
Creo que hasta cierto punto esa realidad ya existe. Quiero decir, una de estas grandes empresas, más grandes, ejecutan toneladas de estas cosas automatizadas todo el tiempo. Así que recopilación de datos automatizada, creación automatizada de nuevos modelos, formas automatizadas de crear nuevos modelos y de encontrar los parámetros óptimos para este modelo. Así que hay mucho de eso ya existe en cierto modo. Pero yendo un paso más allá, digamos que tú, tú quieres… tienes un caso de uso particular en mente, y que quieres crear una IA para ello, así que el nivel será que si tú, tú escribes un comando “créame un modelo de IA que” -no sé- “prediga” -bueno, hablamos de carreras de caballos el otro día, así que por qué no- “prediga el ganador de esta carrera de caballos” algo así, “créame una IA que haga eso”. Posiblemente no esté tan lejos como uno podría pensar, porque estamos hablando de casos de uso bastante estándar, cuando hay un proceso conocido. Creo que eso se puede automatizar hasta cierto punto.

Spanish Virtual Ian Bowie
Bien. ¿Y si lo hacemos muy, muy sencillo? Estoy seguro de que hay un montón de gente por ahí que le gustaría, por ejemplo, vender algo en Internet. Y digamos que alguien tiene una idea. Vamos a mantenerlo realmente simple. Quieren vender bolígrafos en Internet, tal vez de alguna manera pintan a mano los bolígrafos existentes, y luego quieren venderlos en Internet. Así que decidieron, vale, vendo bolígrafos. Quiero saber cómo llegar a mi público. Necesito algún tipo de tienda web para poder venderlos, algún tipo de plataforma de comercio electrónico. Quiero difundirlo en las redes sociales, pero no sé cómo hacerlo. ¿Sería posible tener un sistema de IA que pudiera hacer todo eso por ti?

Spanish Virtual Michael Stormbom
Sí, no se trata sólo de la IA, sino de todo el sistema o la infraestructura que la rodea. Pero, claro, en ese sentido es un proceso predecible, ¿no? Sí. Quiero decir, hablamos… Hay pasos predecibles que tienes que dar ahí. Y que se pueden automatizar.

Spanish Virtual Ian Bowie
El otro día hablamos de los asistentes virtuales. Y ya sabes, tu próximo empleado podría ser un algoritmo. Así que, en cierto modo, esto es lo mismo, ¿no? ¿De verdad? Sí, hoy en día vas a una agencia de publicidad. O ya sabes, hablas con un diseñador web, y les dices lo que quieres, y lo que quieres hacer. Entonces probablemente vas a ver a un experto en medios sociales, y les dices de nuevo, lo que necesitas. Y ellos hacen todo eso por ti.

French Virtual Ian Bowie
Bonjour et bienvenue à AI unfiltered avec moi, Ian Bowie, et notre expert résident : Michael Stormbom. Où nous allons parler de tout ce qui concerne l’IA dans notre société numérique moderne et ce que l’avenir nous réserve à tous.

French Virtual Ian Bowie
J’ai pensé qu’aujourd’hui, il serait intéressant de parler d’un sujet que je viens de lire, à savoir l’IA à faible code ou sans code, et le fait que des personnes non techniques peuvent créer des algorithmes basés sur l’IA sans savoir comment les construire.

French Virtual Michael Stormbom
Oui, en effet, et cela ne s’applique pas seulement à l’IA, mais la règle générale no code, low code, le mouvement, que vous n’avez pas nécessairement besoin de beaucoup de compétences en programmation ou dans le scénario idéal, aucune compétence en programmation que ce soit pour produire, eh bien, des logiciels ou des applications ou en effet l’IA. Donc…

French Virtual Ian Bowie
Ouais, je veux dire, je pense que je trouverais ça incroyablement responsabilisant. Il suffirait d’écrire, d’enregistrer ou de combiner les deux, les paramètres que l’on souhaite définir pour que le programme fasse la magie.

French Virtual Michael Stormbom
Oui, tout à fait. Et bien sûr, une grande partie de l’IA consiste, en quelque sorte, à assembler des pièces de Lego. Il y a ces processus très standardisés, donc même, même la création de l’IA en tant que chose qui pourrait être concevable, je veux dire, dans une certaine mesure, est déjà automatisée aussi.

French Virtual Ian Bowie
Oui, je veux dire, juste par curiosité, à quelle distance pensez-vous que nous sommes de cette réalité ?

French Virtual Michael Stormbom
Je pense que dans une certaine mesure, cette réalité existe déjà. Je veux dire, l’une de ces grandes, grandes entreprises, elles gèrent des tonnes de ces trucs automatisés tout le temps. Donc la collecte automatisée de données, la création automatisée de nouveaux modèles, les moyens automatisés de créer de nouveaux modèles et de trouver les paramètres optimaux pour ce modèle. Donc il y a beaucoup de choses qui existent déjà d’une certaine manière. Mais en allant un peu plus loin, disons que vous voulez… vous avez un cas d’utilisation particulier à l’esprit, et vous voulez créer une IA pour cela, alors le niveau sera que si vous écrivez une commande ” créez-moi un modèle d’IA qui ” – je ne sais pas – ” prédit ” – eh bien, nous avons parlé de courses de chevaux l’autre jour, alors pourquoi pas – ” prédit le gagnant de cette course de chevaux ” genre de chose, ” créez-moi une IA qui fait ça ” ? Peut-être pas aussi loin qu’on pourrait le penser, parce que nous parlons de cas d’utilisation assez standard, surtout lorsqu’il y a un processus connu. Je pense que cela peut être automatisé dans une certaine mesure.

French Virtual Ian Bowie
Ok. Et si on le rendait vraiment, vraiment simple ? Je suis sûr qu’il y a un grand nombre de personnes qui voudraient, par exemple, vendre quelque chose sur Internet. Et donc disons que quelqu’un a eu une idée. Faisons simple. Ils veulent vendre des stylos sur Internet, peut-être qu’ils peignent à la main des stylos existants, et ensuite ils veulent les vendre sur Internet. Alors ils ont décidé, ok, je vends des stylos. Je veux savoir comment atteindre mon public. J’ai besoin d’une sorte de boutique en ligne pour pouvoir les vendre, une sorte de plateforme de commerce électronique. Je veux les diffuser sur les médias sociaux, mais je ne sais pas comment faire. Serait-il possible d’avoir un système d’IA qui pourrait faire tout cela pour vous ?

French Virtual Michael Stormbom
Oui, je veux dire, il ne s’agit pas seulement de l’IA mais de tout le système ou de l’infrastructure qui l’entoure. Mais je veux dire, bien sûr, je veux dire que c’est dans ce sens un processus prévisible, n’est-ce pas. Ouais. Je veux dire, nous avons parlé … Il y a des étapes prévisibles que vous devez prendre ici. Et qui peuvent être automatisées.

French Virtual Ian Bowie
Nous avons parlé l’autre jour des assistants virtuels. Et vous savez, votre prochain employé pourrait être un algorithme. Donc, d’une certaine manière, c’est la même chose, n’est-ce pas ? Vraiment ? Ouais, de nos jours, vous allez dans une agence de publicité. Ou tu sais, tu parles à un web designer, et tu lui dis ce que tu veux, et ce que tu veux faire. Ensuite, vous allez probablement voir un expert en médias sociaux, et vous lui dites à nouveau ce que vous voulez. Et ils font tout ça pour vous.

Ian Bowie
Alright folks, so that actually is going to be the last episode for a few weeks, because we’re going to take our summer break. But we’ll be back with you in August.

Michael Stormbom
Yes, August 30. So until then!

Ian Bowie
We wish you a very happy summer!

Michael Stormbom
Have a nice summer!

Voice-overs in Swedish, Finnish, Norwegian, Danish, Icelandic, Estonian, Latvian and Lithuanian
Ha en trevlig sommar! Mukavaa kesää! Ha en fin sommer! Hav en god sommer! Gleðilegt sumar! Ilusat suve! Jauku vasaru! Geros vasaros!

English parts transcribed by https://otter.ai
Machine translation by https://www.deepl.com
Speech synthesis by https://www.narakeet.com