Skip to main content

Results appear automatically as you type.

DIY Forum
Report

MakoTalk Brings Offline Japanese Speech Synthesis to the ESP32-S3

Reading lead

MakoTalk 2.1.2 is an Arduino library for offline Japanese speech on ESP32-S3: kana text becomes audio over I2S, with a voice dictionary selected and speech triggered by mspeak(). Its default Zundamon voice is built from 123 sound segments. The article notes a constraint: on a 4MB board, the Default partition provides about 1.2MB for the app, less than the roughly 1.56MB sample. Installation is through an Arduino IDE ZIP; commercial use requires prior permission.

Article body

GitHub - airuyoshii-ai/makotalk-arduino: ESP32-S3向け日本語音声合成ライブラリ。かな入力・I2S出力・著作情報付き音声辞書。Arduino IDE対応。
Image credit: Adafruit Blog

MakoTalk, from airuyoshii-ai, is a Japanese speech synthesis library that runs on an ESP32-S3. Text goes in as kana, the Japanese phonetic script, and audio comes out over I2S, the digital audio bus used by small DAC and amplifier boards. A sketch loads the library, picks a voice dictionary, sets the I2S pins, then calls mspeak() to talk. It works without an internet connection and without sending WAV files from a PC.

This is version 2.1.2. The default voice is VOICEVOX’s Zundamon, built from 123 sound segments. The MinimalSpeak example speaks once at startup, and a reset plays it. The library is installed from a ZIP through the Arduino IDE. Voice samples can be heard on the web and aren’t in the download. On a 4MB board, the Default partition gives the app about 1.2MB, so the roughly 1.56MB sample doesn’t fit.

The software terms allow personal study and non-commercial sharing with credit to the author. Commercial use, including evaluation and prototyping, requires asking first, so the project isn’t listed as OSI open source. The voice data carries separate terms, with credit asked for in distributions, writeups and videos.

See more on GitHub.

GitHub

Source: Anne Barela ·Adafruit Blog · 2026-11-17T16:18:08.000Z · Read the original article

Why read this

The article describes a concrete offline ESP32-S3-to-I2S speech workflow and version details, but the voice sample exceeds the stated default partition space and no independent performance test is reported.

AI score 54 · View scoring rationale

Two-model scores assess source reading value, not factual accuracy. Check this editorial article against the original.

  1. gpt-6-luna-2026-09-22 · 57

    The article describes a concrete offline ESP32-S3-to-I2S speech workflow and version details, but the voice sample exceeds the stated default partition space and no independent performance test is reported.

  2. deepseek-flash · 52

    Reports a concrete embedded speech-synthesis library with named voice model, sample size and partition limits, plus non-commercial licensing; value is limited by small scope, no benchmark evidence and unclear hardware path for DIY judgment.

Analyzed: 2026.11.17 16:56 UTC

Arranged / reviewed on Zoowai · 2026.11.17 17:05

Reviewed and arranged by Zoowai editors. Original content and views belong to the source author.

0

Comments & answers 0

Sign in, rate or reply

VISA Mastercard Maestro JCB American Express Diners Club Discover