I Wanted My Computer to Read Japanese to Me. Here's What Actually Worked.

Japanese text to speech has a problem that speakers of alphabetic languages never encounter. Kanji. The character 生 has at least a dozen common readings — なま, せい, しょう, いきる, うまれる, はえる, and more — and the correct one depends entirely on the surrounding context. 生ビール is "nama biiru." 生活 is "seikatsu." 生まれる is "umareru." A TTS engine that picks the wrong reading doesn't just mispronounce a word, it says a completely different word. And then there's pitch accent, which textbooks barely mention but native speakers hear immediately. 橋 (はし, bridge) and 箸 (はし, chopsticks) have the same vowels and consonants — only the pitch pattern differs. Most TTS tools ignore pitch accent entirely and native speakers notice.

I tested every free Japanese TTS tool I could find over the past two weeks. Read the same NHK News article through all of them, plus a chunk of a light novel from Syosetu and a long Japanese Wikipedia article about the Meiji Restoration. Here's what I found.

CastReader handles Japanese better than I expected for a tool that isn't Japan-specific. It's a Chrome extension — install it, open any Japanese web page, click the icon. It reads with Kokoro AI voice while highlighting paragraphs on the actual page. I opened a 2,000-character NHK article about population decline and clicked play. The kanji readings were correct throughout. Pitch accent wasn't perfect — a couple of compound words had flat intonation where a native speaker would have a clear pitch drop — but it was close enough that I kept listening instead of reaching for the stop button. The real value is the paragraph highlighting. When you're reading Japanese, especially if you're not fluent, knowing exactly which sentence is being spoken while looking at the kanji on screen is the difference between passive listening and active learning. My friend Yuki, who's a Japanese teacher in Tokyo, tried it on an Asahi Shimbun article and said the voice was "まあまあ" — which if you know Japanese understatement means she was actually somewhat impressed. CastReader also works on Syosetu, Kakuyomu, Yahoo Japan News, and basically any Japanese website with text. Free, no signup.

VOICEVOX is the open-source darling of the Japanese TTS community. It's a desktop application (Windows, Mac, Linux) that generates Japanese speech with character voices — Zundamon, Shikoku Metan, and about twenty others, each with distinct personality. The voice quality is excellent. Pitch accent is handled better than most commercial options. The catch is that VOICEVOX is designed for content creation, not web reading. You paste text into the application, adjust pronunciation and intonation by mora, then export audio files. There's no browser integration, no web page extraction, no highlighting. If you're making YouTube videos or podcasts in Japanese, VOICEVOX is incredible. If you want to listen to a web article while looking at the page, it's the wrong tool.

Google Translate's Japanese speaker icon is the quickest option for checking how something sounds. Paste Japanese text, click the speaker. Google's neural Japanese voice handles most kanji readings correctly and the rhythm of sentences feels natural. I use it constantly for individual sentences — when I encounter a word in kanji that I know but can't remember the reading for, I paste it into Google Translate and click the icon. Five seconds, done. For full articles the 5,000-character limit is restrictive — a typical newspaper article is around 2,000-3,000 characters in Japanese so one article fits, but anything longer means multiple paste cycles.

Edge Read Aloud has a Japanese voice (Nanami) powered by Microsoft Azure that sounds polished and professional. Open any Japanese page in Edge, Ctrl+Shift+U, and it reads with word-level highlighting. Nanami handles formal Japanese well — news articles, Wikipedia, official documents. On more casual text like blog posts or light novels, the formal tone can feel slightly off, like a newsreader narrating someone's diary. But the pronunciation accuracy is high and the speed control is smooth from 0.5x to 2x. You need Edge, which for most people means maintaining a second browser. On Mac that's less annoying. On Windows where Edge comes pre-installed it might already be there.

macOS and iOS have Japanese voices built in. On Mac — System Settings, Accessibility, Spoken Content, Speak Selection. Download Kyoko or Otoya. Highlight Japanese text anywhere, press Option-Escape. On iPhone, same through Accessibility settings. The voices are older than the AI models but perfectly usable for following along with written text. I use Speak Selection on my Mac when I'm reading Japanese academic papers in Preview — highlight a paragraph I'm stuck on, listen to the reading, and suddenly the sentence structure clicks in a way it didn't from visual parsing alone. Apple doesn't promote these voices much and most Japanese learners I've talked to had no idea they existed.

For reading Japanese web content — news, light novels, Wikipedia, blogs — CastReader in Chrome is where I'd start. The combination of article extraction, paragraph highlighting, and decent kanji reading accuracy makes it the most practical option for people who actually want to listen to Japanese websites. For content creators who need perfect pitch accent control, VOICEVOX is unmatched and free. For quick pronunciation checks, Google Translate is always a paste away. All free. All handle kanji better than anything available five years ago.