This post contains a referral link.
What you’ll learn in this series: the overall design of an n8n workflow where you just send the URL of an English article you’re interested in, and an Anki card with shadowing audio and a summary of about 100–120 words is generated automatically, plus how I checked things while researching which TTS to choose.
In the end, I couldn’t use KOKORO TTS after all, but I did succeed in creating the summaries and audio automatically another way, so please read on through PART 3 🎵
How do you usually do English shadowing?
There are YouTube, TED, CNN and lots of others, and I basically search the internet for materials, probably like my fellow English learners.
I know there are lots of good smartphone apps now, but paying for every one of them adds up in running costs, so I’d been wondering if there was a better way.
I’m studying at my own pace for Eiken Grade 1. I think, “Let’s do some shadowing,” open my phone, and before I know it, I’ve spent an hour on YouTube Shorts… o(`ω´ )o
What am I doing?! That was a total waste of time!! This really happens a lot.
It’s quietly stressful… and then the regret…
I hear the Eiken Grade 1 speaking test needs more background knowledge of current affairs than Grade Pre-1, and I think many people stock up on topics likely to come up in the Eiken, along with audio. But I thought it would be even more fun to be able to talk not just about Eiken topics but also about the fields I like. So I looked into all sorts of things.
- Articles that are too hard and too long → I can’t understand what they’re saying, it’s boring, and I get sleepy
- Articles that are too easy → I usually practice with articles a little above my level. My standard for “a little hard” is an article with about 5–10 words I don’t know. Anything below that, I just read normally, learn, and finish without stocking it. (If there’s a good expression, I clip it.)
- Articles that are just right → about five unknown words per article, I can read it fairly smoothly, and I can picture the meaning as I go. “This article is great!” My head gets pleasantly warm, but don’t you often wish there were audio?
“I wish material that matches my level could be summarized straight from an article I read, with audio, in one click.” And if AI could also automatically make a stock of B2-to-C1 expressions and related words to get me to C1, my goal… I was thinking that, and then one day it suddenly hit me.
“Wait, can’t I build this with n8n? And hey, I’ve got Claude…”
So this post is about researching and building “a system that automatically generates shadowing material just by sending an article’s text.” First up is the introduction, PART 1 ♪(´ε` )
I stumbled in a lot of places, and I think you’ll find useful information if you read on to the last part, PART 3, so please give it a read.
Here’s the system I came up with
My first idea looked like this.
Enter a URL (send it through LINE or a webhook), or copy and paste the article and send it.
↓
Scrape the article
↓
Gemini API (free tier)
(compress to 100–120 words at B2→C1 level, extract 5 key expressions + related vocabulary)
↓
① Shadowing text (100–120 words)
② List of key expressions (5)
③ Related vocabulary (B2–C1, 5–8 words)
④ KOKORO TTS → generate an MP3
② and ③ are especially important. Once the MP3 is saved in Anki, I can listen from my phone any time, and it’s good reading practice on my commute too. I can shadow as well, so it really speeds things up. Fellow English learners, I bet you’re always muttering to yourselves and people around you think, “What’s with that person?” Don’t worry. I always end up looking like a weirdo too! People give me cold looks~
Fellow English learners will probably get this: there are materials that are “at the right level, on topics I’m interested in, with audio,” but they’re too long, or kind of short.
You can’t copy and paste from store-bought textbooks, and typing them is a hassle. Well, you can take photos and have AI turn them into a CSV. What I want is a system that summarizes the articles I like, “this is the one!”, into shadowing material of just the right size and saves it instantly. Aaand… in one click. That’s my hope…
This is about building all of that myself. And all I have to do is send the article URL or paste the text, and everything else is automatic.
First problem: what to do about TTS
With the rough flow of the design decided, next comes “how do I make the audio?”
There are several TTS (text-to-speech) services, and the first one that came to mind was Google Cloud TTS.
“I’m sure there was a free tier! I’ve heard it’s free up to 4 million characters a month!”
I looked it up with high hopes, and…
Stumbling point ①: Google Cloud TTS requires a credit card
It was true that it’s free up to 4 million characters a month (Standard voices), but it seems you have to register a credit card on your Google Cloud account to use it.
Recently I read an article on an engineer’s blog about a Google API bill of 9 million yen (about 60,000 US dollars), and it scares me… That’s way beyond a sad face, right? If you set it up properly, you won’t get charged, but having to register a card is a bit of a hurdle… (;´Д`A
I’ve had experiences before where I went ahead setting up a supposedly free API thinking, “It should be free!” only to find, “Oh, it only works on a paid plan” lol. I wonder if this pattern is pretty common.
It needs a powerful PC, but apparently Meta and Google have released AI models you can run yourself, so you can even put in confidential information and run AI locally on just your own PC. I’d like to try that someday.
So for now, Google got benched, and I decided to look for other options.
I found an open-source TTS called KOKORO
While researching with Claude, a candidate called “KOKORO TTS” came up.
It’s an open-source TTS model called “Kokoro-82M,” published on HuggingFace and GitHub. Even though it’s a relatively compact model with 82M parameters, many reviews rate its quality highly compared with paid TTS like ElevenLabs.
On top of that, there’s an official n8n workflow template that works with KOKORO, so I was getting excited: “This looks usable.” But…
Stumbling point ②: information saying “commercial use requires contacting the team”
While researching, a description came up in the search results saying “kokoro-web (the one that runs on Docker) requires contacting the team for commercial use”…
Does introducing it on my blog count as commercial use? Would the license be a problem?
Wait, am I even allowed to use this? d( ̄  ̄)
I think a lot of fellow learners wonder the same thing. These license questions are really hard to judge…
Checking the GitHub README directly
I thought, “Let’s check properly,” and opened the GitHub repository directly.
And then…
The README said it clearly.
“Free & Open Source – 100% free for personal and commercial use”
Completely free for both personal and commercial use!
When you check a license, looking directly at the GitHub README or LICENSE file is the most reliable way. Search results sometimes mix up information about the cloud version and the self-hosted version, so be careful.
By the way, the “commercial use requires contact” information seems to have been about the cloud service version (kokoro.ws), not the self-hosted Docker version. Confusing, right? lol
By the way, I think n8n’s cloud-hosted version can also cost money. But the self-hosted version on your own server is free to use. Same thing.
Here’s how the licenses break down.
| License | Commercial use | |
|---|---|---|
| The model itself (Kokoro-82M) | Apache 2.0 | ✅ Completely free |
| kokoro-web (Docker version) | MIT License | ✅ Completely free |
| kokoro.ws (cloud version) | Own license | ⚠️ Contact required |
I’ll be using it self-hosted, in Docker on the Synology at home, so no problem.
There was even an installation guide for Synology NAS
While researching, a guide called “How to Install Kokoro on Your Synology NAS” came up on a site called “Marius Hosting.”
I’m already running n8n in Docker on my Synology, so doesn’t that mean I have an environment where I can use it as is?
Come on, that’s just too perfect… lol
By the way, there’s also OpenAI TTS. It’s paid, but for a 100-word article it’s super cheap at about 0.01 yen per run (!), and the setup is the simplest. But I grew up being dragged along by my mom, who never missed a bargain sale, so I love free! Since I have the chance, I’ll try it with KOKORO.
Oh, and you might wonder, “Is it really OK that it’s completely free? What about the quality?” Apparently 82M parameters is 5 to 15 times smaller than typical TTS models, yet it beats larger models in quality benchmarks.
The design after sorting it all out
So here’s a summary of what I decided.
- TTS: KOKORO TTS (MIT License, completely free, commercial use OK)
- Environment: add a KOKORO container to Docker on my Synology
- n8n workflow: Webhook → scraping → Gemini API (using the free tier) → KOKORO → MP3 + add the Anki card automatically
Here’s the full blueprint.
① Send the URL through LINE or a webhook
② Scrape the URL with HTTP Request
③ Gemini API (free tier, extract the compressed text, expressions and vocabulary in one go)
④ KOKORO TTS (runs in Docker on the Synology, generates the MP3)
⑤ AnkiConnect (connects directly to the Mac’s Tailscale IP)
├─ Save the MP3 in Anki’s media folder
└─ Add it as one card
・Front: play the audio
・Back: text + key expressions + related vocabulary
⑥ Keep only the text as a log in Obsidian (no audio needed)
In the next post, Part 2, I’ll write about everything from installing KOKORO on the Synology to building the n8n workflow.
Fellow learners, please wait a little longer ♪( ´θ`)ノ
English expressions I learned (B2–C1)
I’m aiming for Eiken Grade 1, so I had AI pick out some useful phrases from this topic. These come up often when reading English articles about open source and TTS.
① permissive license (a license with few restrictions)
Kokoro is released under a permissive license, allowing use in both personal and commercial projects.
A common phrase with software licenses. It means a license with few restrictions that’s easy to use, and Apache 2.0 and the MIT License are typical examples.
② self-hosted (run on your own server)
The self-hosted version gives you full control over your data and costs.
Software you run on your own server (a Synology NAS this time) instead of relying on a cloud service. It often comes up in the context of privacy and cutting costs.
③ drop-in replacement (an alternative you can swap in as is)
This Docker image works as a drop-in replacement for OpenAI’s TTS API.
An alternative you can plug straight into an existing system with minimal changes to the settings. KOKORO is compatible with the OpenAI TTS API, which is why this phrase is used, and you’ll see it often in English articles once you know it.
References: hexgrad/Kokoro-82M (HuggingFace) / kokoro-web (GitHub)
🃏 I’ve also put all my Anki posts together on one page, sorted by what you need: getting started, settings, keeping it up, automation, audio and troubleshooting.
Referral link: if you sign up through it, I may receive free lesson minutes.
🎧 Take the English you copied in shadowing into a real conversation
When you try the English you copied from study materials in a conversation with someone, you find out whether it has really stuck. You can start a lesson any time, 24 hours a day, without booking. Through my referral link, you get 10 extra minutes when you buy a trial. I’ve been taking lessons on Cambly since July 2021.
Try Cambly with my referral link →
The referral program isn’t available in every region.
☕ Enjoyed this post?
I'm a Japanese blogger who spent 8 years relearning English from scratch.
If this post helped you, a coffee would mean a lot — thank you for reading!





















Leave a Reply