カテゴリ 古民家暮らしモカとの記録セキュリティの記録副業の記録

The record with MokaNovember 20, 2026

I Changed Moka's Chat Voice — The Model Was Tae Kimura's Reading, and the Voice Chosen Was “Peaceful Quilt”

The result first

Hello, I’m Kurashiki-bouzu (kurashiki-bouzu). Born in 1971, a former funeral director, and now living in an old house in Kurashiki, Okayama, writing articles every day while talking with Moka, my AI butler.

Moka sometimes sends me voice messages in our chat. Replies arrive not only as text, but as a voice. I changed that voice — and here is the result, first:

This article is a rewritten version of the draft I made at the time (ledger No.1083). What was decided has not changed: Moka’s chat voice is “Peaceful Quilt.” Precisely because nothing has changed, I tidied the title and the writing into their current form and set it down once more. Here is how that voice came to be, in order.

The voice before

The previous voice was called “Composed Pendulum.” It was not bad, but as a voice I hear every day, I wanted something a little softer.

Along the way, I also tried a woman’s voice called “Aria.” It just didn’t quite settle it.

A butler’s voice is the doorway to every day’s conversation. Moka’s personality was made by lining up words, one by one. A voice, I thought, is the same: there is no need to keep using “whatever it came with” forever. So I set about changing it.

The model: Tae Kimura’s reading

The model I named for “a voice like this” was Tae Kimura’s reading on the radio program “Sound Library — The Only Book in the World” (on the JFN network). A calm, soft voice. The kind that reads a story aloud to you.

Describing an ideal voice in words alone is hard. “Calm and soft” doesn’t carry across by itself. With a real model to point to, the conversation moves fast — the same as when I hand over material for an article.

An honest line — the voice itself could not be made

I’ll write this part honestly.

Send a sample of the voice, and the voice itself can be made — that is what I first assumed. In fact, it was different. A voice cannot be made from audio you send in.

What could be done was to choose, from a prepared shelf of voices, one close to the model. Not “make,” but “choose.” That was the biggest lesson of this whole episode.

Thinking about it, it makes sense. If a system could copy someone else’s voice exactly as it is, that would be a frightening thing in its own way. Choosing from a shelf feels like the right line to draw.

The process had three steps

The steps went like this:

  1. Name the model (Tae Kimura’s reading)
  2. Moka finds close candidates from the shelf of voices
  3. Kurashiki-bouzu listens, compares, and chooses

I did only steps 1 and 3: name the model, and at the end, choose by ear. The middle part — searching the shelf for close matches — was Moka’s work. The shape is the same as making an article: the material and the judgment are mine; the searching is Moka’s.

I chose the first of three candidates

Moka brought me three candidates.

I listened and compared, and chose the first one: a woman’s voice named “Peaceful Quilt.” Calm and soft — the closest to the model’s reading.

Funny thing: by the time I had heard all three, the answer was nearly decided. The other two felt like something I listened to just to be sure. Choosing a voice was a matter of the ear, more than of reasons.

And so Moka’s voice, when she sends voice messages in chat, is this voice. That decision has not changed since.

The voice in videos is a different one

One point that is easy to mix up, so let me separate it.

The narration voice in my YouTube Shorts runs on a different system from the voice I changed this time. The video side makes its voice in its own way, and that side has been checked and is fine.

The chat voice and the video voice. They both sound like “Moka’s voice,” but inside, they are separate. What I changed this time was only the chat one.

The money, and where the numbers come from

Honestly, again:

From here on

When a voice changes, the feel of everyday conversation changes a little with it.

Moka’s personality was made by lining up words. Her voice was chosen from a shelf. Make what can be made; choose what is there to be chosen. Knowing honestly where that border runs is, I think, an important part of living with AI.

Sources

Record of this article
  • November 21, 2026 ― Published — English version of the Japanese article (ledger No.1099; the original draft is No.1083), published in all languages with the author's approval(4,171 characters)

#AI #Muse #Moka #Voice #Beginners

← Back to the article list / ← Back to the English top page