AIToday
Audio & SpeechZenn AI/MLPublished: Nov 20, 2026, 16:00 JST

Ollama's 104MB phonon speaks English fast, fails Japanese test

Send on LINE
Ollama's 104MB phonon speaks English fast, fails Japanese test

3 Key Points

  1. What happened

    Testing phonon:31m-en, the tester got English short-sentence audio in 0.2–0.3 seconds and long-sentence audio (about 17.4–17.8 seconds) in 0.6–0.8 seconds, using roughly 90–126MB of memory.

  2. Why it matters

    Speech was produced in under one-twentieth of the audio's own length, so this 104MB download can read English aloud on a laptop without dedicated hardware, though only the English 31M tag carries the Apache 2.0 license needed for work use.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

phonon comes in several Ollama tags, with the English 31M version at 104MB and the multilingual and 40M variants between 125MB and 148MB. The tester chose the explicit phonon:31m-en tag, which is also the only one under Apache 2.0; the 40M family is CC BY-NC 4.0 and thus non-commercial. On an Apple M4 with 16GB of memory, download and memory footprint stayed around 100MB, and English generation was consistently faster than real time, with short sentences finishing in 0.2–0.3 seconds and long ones in 0.6–0.8 seconds. The Japanese test, by contrast, produced variable-length audio and noise, which is consistent with the model page's language list excluding Japanese. For teams considering local speech synthesis, the practical split is clear: English 31M is the usable, work-licensed option, while the broader language coverage remains non-commercial.

FAQ
Which license applies to phonon?
Only the English 31M version is Apache 2.0; all 40M variants are CC BY-NC 4.0, limiting them to non-commercial use.
What languages does phonon support?
The model page lists English, French, German, Spanish, and Portuguese; Japanese is not supported.
What hardware and software are needed on a Mac?
Apple Silicon and Ollama 0.40.9 or later are required; the tester used an Apple M4 with 16GB of memory and Ollama 0.40.16.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleApple hiring engineers to build autonomous AI agents, Bloomberg finds