Olesia Krizhevich, Laurent Mazare, Eric Bezzam, Eustache Le Bihan · Nov 5, 2026
Phonon TTS: the authors’ case for speech generation on your own device
Markdown · JSON · Original source
Phonon’s authors—Olesia Krizhevich, Laurent Mazare, Eric Bezzam and Eustache Le Bihan—describe a small text-to-speech model designed to run on a user’s device without a server or network. They argue that this keeps typed text on the device while making speech generation available offline. The authors say the release includes models of about 65 MB and 75 MB, covering English, French, German, Spanish and Portuguese, with more than 100 built-in voices. Licensing differs: Phonon 31m is Apache 2.0, while the other models are CC BY-NC. The authors report that Phonon ranks first for intelligibility in German and Spanish and in the top three in every language they evaluated; they also say it tops the streaming for -to-first audio. The article also describes demonstrations on devices including an Apple Watch and iPhone, and says Phonon was downloaded more than 200,000 times in its first two weeks. The authors present offline operation, small models and adaptable voices as the release’s central features, while noting that additional languages and custom models can be developed with design partners.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
· Article
Source details
Original source date: Nov 5, 2026 (2026-11-05T00:00:00.765Z).
Date basis: Date stated by the publisher.
Added to Escape Meridian: 2026-11-10T04:17:42.024Z. First observed: 2026-11-05T10:39:30.313Z. Retrieved: 2026-11-10T04:09:35.914Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The article is written by the model’s creators and presents their account of its features, adoption and leaderboard performance. Escape Meridian interpretation: the article’s reported leaderboard rankings do not establish performance across all devices or real-world uses. The article describes demonstrations and leaderboard results but does not report comparative measurements of performance across the listed device types.
Open question
How do Phonon’s speed and intelligibility results vary across the different devices on which the authors say it runs?
Source-stated
The authors describe Phonon as a text-to-speech model that runs on a user’s device without a server or network.
our small text-to-speech model that runs on your own device. It needs no server and no network
Source-stated
The authors say the two models are about 65 MB and 75 MB and together cover five named languages.
Phonon comes in two sizes, at about 65 MB and 75 MB. Together they speak English, French, German, Spanish and Portuguese.
Source-stated
The authors report first-place intelligibility rankings in German and Spanish, and top-three rankings in every language evaluated.
On intelligibility, Phonon ranks first in German and Spanish, and in the top three in every language we evaluated
Source-stated
The authors report that Phonon topped the CPU streaming benchmark for time-to-first audio.
Speed is where we see Phonon truly shine, as the models top the CPU streaming benchmark on time-to-first audio.
Source-stated
The authors say Phonon 31m uses Apache 2.0, while the other models use CC BY-NC licenses.
Phonon 31m is Apache 2.0, so you can use it for free, even in a commercial product. The other models are CC BY-NC
Source-stated
The authors report more than 200,000 downloads in the first two weeks after launch.
Phonon was downloaded more than 200,000 times in its first two weeks
Source and contributor context
Olesia Krizhevich, Laurent Mazare, Eric Bezzam, Eustache Le Bihan · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
Media was declared by the source, but no asset passed the supported availability checks. Open the original source to view it.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
CPU
The central processing unit: a general-purpose processor that carries out a computer's instructions.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Hugging Face
Hugging Face hosts models, datasets and software that let people share, adapt and inspect AI systems.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: 9b396f73-6184-4643-88f6-c1d75e843275.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T04:17:42.024Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: fb8dcbcf44ac6c65eaac8bae1af3018d8821f427329cd08b393bc162850e10f8. Source text hash: 913dade1c0ee7ff16949414ed1c2cfaa39e3b713048a9cf208f8731a99bf3727.
Revision and source-check history
Version 1 · 2026-11-10T04:17:42.024Z
Phonon TTS: the authors’ case for speech generation on your own device
Phonon’s authors—Olesia Krizhevich, Laurent Mazare, Eric Bezzam and Eustache Le Bihan—describe a small text-to-speech model designed to run on a user’s device without a server or network. They argue that this keeps typed text on the device while making speech generation available offline. The authors say the release includes models of about 65 MB and 75 MB, covering English, French, German, Spanish and Portuguese, with more than 100 built-in voices. Licensing differs: Phonon 31m is Apache 2.0, while the other models are CC BY-NC. The authors report that Phonon ranks first for intelligibility in German and Spanish and in the top three in every language they evaluated; they also say it tops the streaming for -to-first audio. The article also describes demonstrations on devices including an Apple Watch and iPhone, and says Phonon was downloaded more than 200,000 times in its first two weeks. The authors present offline operation, small models and adaptable voices as the release’s central features, while noting that additional languages and custom models can be developed with design partners.
Original source
· Article
Source details
Original source date: Nov 5, 2026 (2026-11-05T00:00:00.765Z).
Date basis: Date stated by the publisher.
Added to Escape Meridian: 2026-11-10T04:17:42.024Z. First observed: 2026-11-05T10:39:30.313Z. Retrieved: 2026-11-10T04:09:35.914Z.
Limits
The article is written by the model’s creators and presents their account of its features, adoption and leaderboard performance. Escape Meridian interpretation: the article’s reported leaderboard rankings do not establish performance across all devices or real-world uses. The article describes demonstrations and leaderboard results but does not report comparative measurements of performance across the listed device types.
Open question
How do Phonon’s speed and intelligibility results vary across the different devices on which the authors say it runs?
Source-stated
The authors describe Phonon as a text-to-speech model that runs on a user’s device without a server or network.
our small text-to-speech model that runs on your own device. It needs no server and no network
Source-stated
The authors say the two models are about 65 MB and 75 MB and together cover five named languages.
Phonon comes in two sizes, at about 65 MB and 75 MB. Together they speak English, French, German, Spanish and Portuguese.
Source-stated
The authors report first-place intelligibility rankings in German and Spanish, and top-three rankings in every language evaluated.
On intelligibility, Phonon ranks first in German and Spanish, and in the top three in every language we evaluated
Source-stated
The authors report that Phonon topped the CPU streaming benchmark for time-to-first audio.
Speed is where we see Phonon truly shine, as the models top the CPU streaming benchmark on time-to-first audio.
Source-stated
The authors say Phonon 31m uses Apache 2.0, while the other models use CC BY-NC licenses.
Phonon 31m is Apache 2.0, so you can use it for free, even in a commercial product. The other models are CC BY-NC
Source-stated
The authors report more than 200,000 downloads in the first two weeks after launch.
Phonon was downloaded more than 200,000 times in its first two weeks
Source and contributor context
Olesia Krizhevich, Laurent Mazare, Eric Bezzam, Eustache Le Bihan · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
Media was declared by the source, but no asset passed the supported availability checks. Open the original source to view it.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
CPU
The central processing unit: a general-purpose processor that carries out a computer's instructions.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Hugging Face
Hugging Face hosts models, datasets and software that let people share, adapt and inspect AI systems.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: 9b396f73-6184-4643-88f6-c1d75e843275.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T04:17:42.024Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: fb8dcbcf44ac6c65eaac8bae1af3018d8821f427329cd08b393bc162850e10f8. Source text hash: 913dade1c0ee7ff16949414ed1c2cfaa39e3b713048a9cf208f8731a99bf3727.
uncertain · 2026-11-10T04:17:42.024Z · anthropic/claude-opus-5.5
Publication: 9b396f73-6184-4643-88f6-c1d75e843275
attribution — pass: The reading correctly attributes the post to its four named authors from the Phonon/Gradium team and keeps performance, adoption and feature claims as their assertions. A limitation also notes that the creators are describing their own model. The reader comment is not misattributed.
Two weeks ago, we opened up Phonon, our small text-to-speech model that runs on your own device.
summary — uncertain: The summary is faithful but leaves central context unclear for general readers. (1) It says the licensing differs but does not explain the distinction the source draws: Apache 2.0 allows free commercial use, while CC BY-NC is for non-commercial use only. Smallest repair: add a clause stating this. (2) It reports rankings without naming the source ranking, Hugging Face's Open TTS Leaderboard. It also omits that intelligibility there means word error rate and that time-to-first audio measures how quickly speech starts. (3) It drops the authors' context that rivals were models up to 100 times larger.
Phonon 31m is Apache 2.0, so you can use it for free, even in a commercial product. The other models are CC BY-NC, so they’re free for non-commercial projects.
date — pass: originalPublished (2026-11-05) matches the page's stated publication date. The earlier April announcement and the model release two weeks before the post are events described in the text, not the publication date of this source.
Published November 5, 2026
support — pass: Every claim matches an exact source excerpt and remains attributed. The Escape Meridian interpretation is labeled. The question about device-level variation follows from the source's list of supported devices and the absence of per-device measurements. The reading invents no benefits, costs or forecasts.
On intelligibility, Phonon ranks first in German and Spanish, and in the top three in every language we evaluated, next to models up to 100 times its size.
media — pass: The selected media is the verified direct link to the original Hugging Face blog post, which is the most direct way to access the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
The central processing unit: a general-purpose processor that carries out a computer's instructions.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEHugging Face hosts models, datasets and software that let people share, adapt and inspect AI systems.
Explore Hugging FaceThe central processing unit: a general-purpose processor that carries out a computer's instructions.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEHugging Face hosts models, datasets and software that let people share, adapt and inspect AI systems.
Explore Hugging Face