作業の説明とプロセス
Hello, I'm TakeNom. Over the past few days, I've been working on translating audio content into English, and I spent the weekend having fun.
Last night, I created an illustration diagram showing the solution workflow, but there's still a lot of manual work involved.
Right now, I'm speaking on StandFM, talking directly into my smartphone.
The RSS linking is incredibly convenient. It automatically connects to Spotify, automatically links to Listen, a transcription service, and automatically connects to Apple Podcasts as well.
These automatic connections work perfectly without requiring any manual effort.
The red arrows in my diagram represent the manual steps.
Listen transcribes the audio to Japanese text, but after that, quite a bit of manual work is needed.
I use Claude, or any translation app or AI service, to translate the Japanese text into English.
The translation itself isn't that difficult, but then I use Edge's automatic text-to-speech feature to read the translated English text aloud.
This process takes time. I have to play the audio and record it again in StandFM.
While the original channels, Spotify, Listen, Apple Podcasts, receive automatic updates, I've set up a separate Spotify account for English content.
I considered creating a separate StandFM account for RSS purposes, but account switching on smartphones is quite cumbersome.
音声認識とアバターの実験
You need to log out and log back in.
So instead, I upload the English audio to my existing StandFM account, manually download it, and then upload it to the English Spotify channel.
So if I speak for 10 minutes, the translation process is quick, but recording the English text-to-speech takes another 10 minutes, real time.
The downloading and uploading process isn't too time-consuming, but this seems to be the limit of what's possible with the current workflow.
There are some services that can read text in my own voice, but they're mostly paid and seem complicated, so I haven't explored them yet.
I also experimented yesterday with creating an avatar based on my photo that speaks my words.
It works reasonably well, it doesn't look like actual video footage of me speaking, but it's an avatar that resembles me speaking these lines.
I haven't expanded into this area much, but the more I explore, the deeper it gets.
There are so many services available that it's hard to know which is the best practice.
I usually just try the top search results on Google, assuming they're the more popular options.
If it becomes possible to have my own voice read the content with an avatar that looks like me speaking, things could change dramatically.
Right now, humans are still speaking with originality, but AI can apparently analyze a person's way of thinking and speaking patterns to generate content that sounds like them.
With deep learning and other technologies, it could almost seem like I'm speaking, which leads to the extreme possibility that this could function without human presence at all.
アニメの影響とテーマ
Serial Experiments Lane influenced changing topics.There used to be an anime called Serial Experiments Lane.
It came out when I was in high school, and while it was somewhat obscure, it was a psychological horror type of show.
This was during the early days of the internet, when BBSes and chat rooms were just starting.
The anime featured smartphones and similar gadgets, which wasn't uncommon, but Serial Experiments Lane was particularly forward-thinking in its content.
The show explored the concept that even without a physical body, one could exist as a concept in what they called the wired, not the internet.
The worldview suggested that as long as you exist as a concept online, you can connect with anyone from anywhere.
It was somewhat frightening, but I think it's a work that still holds up today.
It was quite popular overseas with a core fan base, and it's been about 20 years now, actually more than that.
Last night, I learned that while the IP isn't free, it's possible to create works with permission.
A kind of Serial Experiments Lane fan-made game is apparently being released on Steam, which piqued my interest.
Since it's on Steam, I could play it on PC, so I'm thinking about buying it.
There was also a PlayStation version of the game, which was quite scary and created an eerie feeling.
Now, it probably has a premium price of over ¥10,000, but back then it was around ¥6,000.
It was quite an advanced game for addressing internet topics, also exploring the concept of existing without a physical body.
AI generation and human value by combining these solutions, we're creating something like the world of Serial Experiments Lane.
Almost everyone has a smartphone now, and there are stories about how smartphones have high penetration rates even in developing countries where home infrastructure isn't well established.
Generative AI is on the rise, and while translation from Japanese to English has been possible with Google Translate, now AI can generate text that resembles someone's personality.
I haven't gone that far yet, but using Microsoft Edge, which Windows users are familiar with, makes it possible to deliver English audio.
Eventually, this will become fully automated in multiple languages, removing the need for all this manual work.
生成的AIと人間の独自性
So the world of Serial Experiments Lane is very much alive now, or perhaps even more so with Generative AI.
What I find somewhat scary is wondering where our originality will hold meaning and value in the future.
In terms of knowledge, humans can't compete anymore.
AI has learned from all kinds of papers and academic theories, and human knowledge pales in comparison.
This raises questions about the fundamental value of humans and what we'll need to seek out in the future.
Maybe it's something like enlightenment in Buddhism? The answer might lie in that realm.
This weekend, as I played around with different tools, I started to see the possibilities more clearly, so I wanted to share these thoughts.
Thank you for listening.
