Last reviewed August 1, 2026
Transparency, testing, and model sources
OfflineTTS combines a working browser application with technical guides about local speech tools. This page identifies who is responsible for the site, what we test ourselves, what comes from an upstream source, and how we correct mistakes.
Operator and responsibility
OfflineTTS is operated and maintained by the OfflineTTS Project Team. Product releases, site maintenance, privacy documentation, and editorial corrections are owned by this team rather than by any upstream model provider.
The OfflineTTS application is proprietary. Its source code and internal development history are not published. This is separate from the licenses of the upstream speech models and runtimes listed below. Questions, reproducible bug reports, licensing concerns, and correction requests can be sent to [email protected].
“OfflineTTS Editorial Team” is the collective byline used on articles maintained by the project team. It is not a claim that an unnamed outside expert reviewed every article. Published and updated dates appear on each article so readers can judge how current it is.
Testing method
We separate first-party observations from upstream specifications and third-party benchmark results. A model size, language count, price, or benchmark score attributed to another project should link to that project's documentation or original dataset. A browser-behavior claim should be reproducible with the steps below.
Reference environment
- Recorded: August 1, 2026
- Operating system: macOS 26.6
- Browser: Google Chrome 150.0.7871.187
- Build checks: Node test runner plus the production Astro build
What the release check covers
- Canonical URLs, robots directives, redirects, and sitemap membership
- Static rendering of product explanations, sources, and dates
- Tool route availability and production asset generation
- Disclosure links and navigation to support information
Repeatable speech-tool checks
- Record the engine, model variant, voice, language, browser version, operating system, backend, and test date.
- Use fixed text containing short sentences, punctuation, numbers, names, and a longer paragraph. Multilingual tests use native-language text rather than machine-translated English fragments.
- Test the first model load and a cached repeat visit separately. Offline claims are checked only after the required assets have been downloaded.
- Confirm that generated audio plays and that exported WAV or MP3 files open outside the page. Transcription checks retain the original clip for comparison.
- Inspect browser network activity when making privacy claims. Kokoro non-English phonemization is documented separately because text is sent to the phonemization service while audio synthesis remains local.
- Report limitations and failed cases. We do not publish a first-party numeric voice-quality score unless the test corpus, comparison method, sample count, and date are also disclosed.
Editorial and correction process
Product documentation, model cards, source repositories, and original benchmark datasets are preferred over summaries. Commercial comparisons distinguish published provider facts from our interpretation. Affiliate relationships do not change access to OfflineTTS and are disclosed next to paid links and on the affiliate disclosure page.
An update date means the page was materially checked or changed; it is not automatically rewritten to make it appear fresh. If a factual claim cannot be supported, the preferred correction is to remove or qualify it rather than silently replace it with another unsupported claim.
Model sources and licenses
OfflineTTS integrates third-party models and runtimes; it did not train the upstream foundation models. The table records the primary source and software/model license presented by the upstream project when this page was reviewed. Individual voice datasets or replacement model files may have separate terms.
| Integration | Primary source | Recorded license | Important qualification |
|---|---|---|---|
| Kokoro-82M | Hugging Face model card | Apache-2.0 | Browser inference uses kokoro-js; non-English text also uses the documented phonemization service. |
| Kitten TTS | KittenML repository | Apache-2.0 | Check the exact model release and bundled notices when redistributing model files. |
| Piper | Rhasspy Piper archive | MIT for the referenced software | Piper voice models originate from multiple datasets. Review the model card and dataset terms for the particular voice you use. |
| Supertonic | Supertone repository | MIT | The browser downloads the public ONNX assets and performs synthesis locally after download. |
| Whisper | OpenAI Whisper repository | MIT | Browser execution also depends on Transformers.js, ONNX Runtime Web, and the browser-whisper integration. |
This summary is not legal advice. Rights in source recordings, text, characters, trademarks, and a specific voice dataset remain separate from the software license.
Public change log
August 1, 2026
Published this transparency record, added visible editorial responsibility, and removed supporting category, directory, and experimental voice pages from the indexable sitemap inventory.
July 28, 2026
Consolidated overlapping search landing pages and pointed internal links directly to their canonical destinations.
July 27, 2026
Replaced individual voice landing pages with one functional voice catalog and direct voice selection in the browser tool.
This page records material public-facing changes. Internal implementation records and proprietary source history are not published.
Corrections and feedback
Send the page URL, the exact statement or behavior in question, the evidence you used, and—when reporting a tool problem—your browser, operating system, selected engine, and reproduction steps to [email protected]. The OfflineTTS Project Team reviews supported corrections and records material public changes through page dates or the change log above.