Corrections and change log
This page records factual corrections and material changes. Verification details, including expected versus actual values and source URLs, are kept in the project repository at docs/verification-log.md.
Duplicate tool pages merged
Six tool pages that opened the same generator with different titles (video to text, caption generator, transcript generator, speech to text, offline and private subtitle generator) were consolidated into /video-to-srt, /audio-to-srt and the home generator. Old addresses return a 308 redirect and the static pages carry a fallback redirect. The homepage also lost a repeated file-size FAQ; one answer now covers both phrasings.
Model download sizes re-measured
Encoder and decoder ONNX file sizes were re-read from the Hugging Face API for whisper-tiny, whisper-base and whisper-small. The UI now shows WASM sizes of about 41 MB, 77 MB and 249 MB. The source URLs and method are in docs/verification-log.md.
Language list checked against the tokenizer
All 99 language codes were extracted from the multilingual Whisper tokenizer and vendored in src/data/whisper-language-codes.json. A unit test fails if the site catalogue and the vendored code list diverge.
No site-wide COOP/COEP
Cross-Origin-Embedder-Policy breaks AdSense frames. The default tool uses the single-thread ffmpeg.wasm build, which needs no isolation. Isolation headers are scoped to /studio only, and /studio carries no ads.
Language pages excluded until accuracy is tested
Language pages carry noindex until a real clip in that language has been transcribed and recorded in the verification log. The visible accuracy note explains this; the tool remains usable.
Initial publication
VoiceToTextFree launched with the browser-based generator, SRT/VTT/TXT exports and the first subtitle guides.
Found an error? Send a correction with the page URL and a primary source.