v0.6.4

Which file do I download?

file	who it is for	what it does
transcript-normalizer-0.6.4-<system>-lite	anyone who just wants to use it	fetches the platform's captions and fixes domain terms; opens a menu when double-clicked
transcript-normalizer-0.6.4-<system>-full (macOS, Linux)	anyone who also has videos without captions, or audio files	everything lite does, and also transcribes audio on your computer (the first time, it downloads a 1.5 GB speech model). On macOS, a .zip: unzip, then open transcript-normalizer inside the folder. On Windows, full is the two files below
transcript-normalizer-0.6.4-windows-full-setup.exe	Windows, full: the installer	choose where it goes (for you only, no administrator needed, or for all users); Start menu shortcut; clean uninstall from "Add or remove programs", which keeps your files
transcript-normalizer-0.6.4-windows-full-portable.zip	Windows, full: portable	unzip anywhere, e.g. a USB drive, and open transcript-normalizer.exe inside the folder; everything (your files, settings, the speech model) stays in its data folder; delete the folder to remove it
transcript_normalizer-0.6.4-py3-none-any.whl, .tar.gz	Python developers	the package, for pip install

<system> is windows, macos (Apple silicon) or linux; Linux executables run on glibc 2.35 or newer (Ubuntu 22.04 and later). The executables are not signed, so Windows and macOS will warn the first time; the SHA-256 of every file is at the end of this page.

Qual arquivo baixar?

arquivo	para quem	o que faz
transcript-normalizer-0.6.4-<sistema>-lite	quem só quer usar	pega a legenda da plataforma e corrige os termos da área; abre um menu com dois cliques
transcript-normalizer-0.6.4-<sistema>-full (macOS, Linux)	quem também tem vídeos sem legenda, ou arquivos de áudio	tudo o que o lite faz, e ainda transcreve o áudio no seu computador (na primeira vez, baixa um modelo de fala de 1,5 GB). No macOS, um .zip: descompacte e abra o transcript-normalizer dentro da pasta. No Windows, o full são os dois arquivos abaixo
transcript-normalizer-0.6.4-windows-full-setup.exe	Windows, full: o instalador	você escolhe onde instalar (só para você, sem administrador, ou para todos os usuários); atalho no menu Iniciar; desinstalação limpa em "Adicionar ou remover programas", que mantém os seus arquivos
transcript-normalizer-0.6.4-windows-full-portable.zip	Windows, full: portátil	descompacte em qualquer lugar, por exemplo num pendrive, e abra o transcript-normalizer.exe dentro da pasta; tudo (seus arquivos, configurações, o modelo de fala) fica na pasta data dele; apague a pasta para removê-lo
transcript_normalizer-0.6.4-py3-none-any.whl, .tar.gz	quem programa em Python	o pacote, para pip install

<sistema> é windows, macos (Apple Silicon) ou linux; os executáveis de Linux rodam com glibc 2.35 ou mais nova (Ubuntu 22.04 em diante). Os executáveis não são assinados, então o Windows e o macOS avisam na primeira vez; o SHA-256 de cada arquivo está no fim desta página.

---

YouTube's bot check said plainly, and a YouTube sign-in for it (D-071); normalizing about three times faster with identical results; the menu no longer waits for the network at startup.

• On Windows, pack propose --whole wrote the contribution file with Windows line endings, so it was not the pack file byte for byte; it is now copied as it is.
• Test builds: the release workflow can be started by hand (Actions → release → Run workflow) on any commit. It builds and tests the same executables, named after the commit, and keeps them as workflow artifacts <sha>-<os>-<flavor>; no tag check, no Release. Tag pushes release as before.
• --gold-draft, the old name of --corrections, is removed (D-023, amended); it was kept far longer than its one release.
• Docs: the pip install lines point at v0.6.3 instead of v0.4.2, whose ingest extra lacked D-058's faster-whisper/av bounds. GUIDE §8 and the README's Status give the current measured numbers and name scripts/measure.py as their source. The ask band (D-060) is documented. Smaller fixes to sample output, the merge backup name, the phonetic classes in CONTRIBUTING and the release page's Windows full row.
• Residue removed after the 2026-10 inspection (docs/INSPECTION-2026-10.md): the unused contains_words and its test, META_FIELDS, evaluate's two one-line wrappers; fits() is now what normalize and measure.py call. Tests share one bundled-pack fixture and cache the fixture's medium band, which they used to recompute per call.
• Tests pin three matcher rules no test covered: D-047's ranking (a canonical term at 80 beats a variant at 84 for the same span) and the D-040 and D-049 guards, before and after overlap resolution, including the case that shows why they are needed (a rejection that removes the longer name). Matching is unchanged.
• Normalizing is about three times faster, with byte-identical results: each caption window is compared only with the pack forms that can match it (its exact equals, and fuzzy forms within two letters of its length, D-005), and the phonetic pass no longer checks every window against every applied change, which made it quadratic in caption length. wxgFO_fyfXg (40 min): 2.29 s → 0.69 s; the same caption four times over: 13.2 s → 3.2 s.
• The menu never waits for the packs repository at startup (D-069, amended). It looks for pack updates in the background, keeps the index for a day in packs/.index-cache.json (inside the folder in portable mode), and when a copy of yours can be merged it says so in one line and offers "Merge an update" in Packs, instead of asking "Merge?" on the first screen. Offline with a copy, the first screen went from 3.2 s to 0.2 s.
• YouTube's bot check is said plainly (D-071). Instead of yt-dlp's raw "Sign in to confirm you're not a bot. Use --cookies-from-browser or --cookies", fetch says "YouTube asked to confirm you are not a bot. This depends on your connection, not on the video." (and, for an age check or a members-only video, that YouTube asks for a signed-in account), stops, and exits 3. The menu then offers: use your browser's YouTube sign-in (cookies), try again later, use a file from this computer, or back.
• YouTube sign-in (D-071): Settings → YouTube sign-in takes a browser (Firefox first; Chrome and Edge on Windows usually cannot be read), a cookies.txt file, or none. fetch --cookies-from-browser <browser> and fetch --cookies <file> do the same for one run and win over Settings; in portable mode a cookies.txt in data/ is used without asking. config.toml keeps only the browser's name or the file's path; yt-dlp reads the cookies itself, and no file the tool writes holds one. A browser that cannot be read is named, with Firefox or a cookies.txt as what to try. Help and both guides have a "YouTube says I'm a bot" section with how to export cookies.txt.

SHA-256

bc38f92dae102a01b20d85a67c6cc888a02090fe4fc1fe97db14ede8a7f11c71  transcript-normalizer-0.6.4-linux-full.tar.gz
5914547c5287ec79a43774efc9e20467860b7932ff36941299714bf7a586c894  transcript-normalizer-0.6.4-linux-lite.tar.gz
810bf1dacbd1cf8b004c34d16b247c86afe468e12a31e47fd7117fb7fbba51be  transcript-normalizer-0.6.4-macos-full.zip
5aaa9cbfebf12bf75aef1866cefbf59e9db364026f36b9db68caca180c0efcfe  transcript-normalizer-0.6.4-macos-lite.zip
4355cc1f85a5a5cf498afd137909855d305df36b1744c9467ade742c1b015a9b  transcript-normalizer-0.6.4-windows-full-portable.zip
4dc9476b2b25e1ed104165105284c31957d1a4a0e6fbda3bbe04002158ee7289  transcript-normalizer-0.6.4-windows-full-setup.exe
618cad2e3f244b40320f364cc242b2c1717355fb74becc371d3d766619548468  transcript-normalizer-0.6.4-windows-lite.exe
72874d66475e2a55b530d5f545bc7a7cbe7b1132fa7dbb0282d4188929a82b50  transcript_normalizer-0.6.4-py3-none-any.whl
a8839f9eeaf6de0662a148a36554c30e5a6e35c0317a76d9598489239b34f99c  transcript_normalizer-0.6.4.tar.gz