Changelog

0.1.4 — 12 October 2026

New in Pro: Teach Rostra a Word

• Fix any word, name, brand or term in a few seconds, and Rostra remembers it. When Rostra says a word wrong, select it, choose Teach Rostra a Word… from the menu, and type how it sounds: hyphens between syllables, capitals on the stressed part ("jah-LELL"). Press Hear it to check, adjust by ear, and save. Or press the microphone and say the word once. From then on Rostra uses your reading everywhere, including possessives ("Jalel's"). Single words and whole phrases both work. Your list lives under Pronunciations in the menu, where you can edit it, and export it to take to another computer. The microphone is only used while you press it, and nothing leaves your computer. On Mac and Windows.

Reading

• Short names that start with a vowel. A name read on its own, like Anna, Olga or Emma, could pick up a sound that isn't there ("Vanna", "Vulgar"). It now starts on the right sound, and longer names keep their full ending.
• "Live" and "As" in more sentences. Proper Pronunciations fixed.

Changed

• A tidier menu. Teach Rostra a Word and your Pronunciations sit together, and the engine status line has moved further down the menu, where it is easier to read.

0.1.3 — 23 September 2026

Added

• An Intel Mac build. Rostra has been Apple Silicon only. There is now an Intel Mac download alongside Apple Silicon and Windows.

Reading

• Technical writing reads the way people say it. Licences and versions: "GPL-3.0" and "AGPLv3" read as "G P L version three point zero" and "A G P L version three", and "IPv6" as "I P version six". File names and abbreviations no longer break the sentence at every dot — "config.json", "README.md", "U.S.", "Ph.D.", "bbc.com". Lowercase acronyms such as "cli", "ci" and "ux" are spelled out instead of being said as invented words. Numbers inside names read as engineers say them: "SHA-256" is "two five six", "Python 3.12" is "three point twelve".
• Roman numerals read by context. "Stage two", "Henry the Eighth", "Super Bowl fifty-eight", "the twenty-first century" — and an IV drip is no longer "Roman four". A numbered list reads as one, two, three.
• Numbers around a colon. "16:9" reads "sixteen by nine" and "3:1" reads "three to one", while "9:30" stays a time. Bible references such as John 3:16 read as verses without turning on the Scripture preset.
• Words that change with meaning. "Record", "content" and "contract" take the right reading in "a new record" and "the content team", and "go live" and "confirmed live" use the broadcast sound. "I read the file" is spoken in the past tense when the sentence makes that clear.
• Ordinary sentences. A number range no longer damages the word before it, so "takes 5–10 minutes" reads as written. A closing quotation mark after a number is no longer read as inches. Two lines running together are no longer read as a web address.

Fixed

• The floating player can be moved again on macOS 27. After the macOS 27 upgrade the player stayed where it appeared and could not be dragged. It moves normally again, and its buttons still work when pressed.

Changed

• Better diagnosis when something goes wrong. Rostra now records more about each read, so a problem reported after the fact can be traced instead of guessed at.

0.1.2 — 26 August 2026

Added

• Units are read as units, not spelled out. "15 MW", "12 µm", "10 kΩ" and thousands of other combinations are now spoken correctly without each one having to be added by hand — where before "kW" worked and "MW" did not. Heights and sizes written with prime marks are read as measurements at last — 5'11" is "five feet eleven inches" and 12" is "twelve inches", where every other text-to-speech tool we tested reads them as silent pauses. Coordinates come through as coordinates, and quoted text is left alone. Throughput and rates read the way they are meant: "35 tok/s" is tokens per second, "72 bpm" is beats per minute, "30 psi" is pounds per square inch, and a data rate now tells bits from bytes correctly. Dimensions read as dimensions — "9x13" is "nine by thirteen". Compound rates read properly too, ranges say "five to ten kilometres" instead of running the numbers together, blood pressure reads "one twenty over eighty millimetres of mercury", and "1/2 cup" is "half a cup". Compass bearings stay bearings: 95°W is west, while 60 W is watts.

Changed

• Rostra's log file no longer grows without limit. It is now capped, keeping recent history and discarding the oldest, so a long-running install cannot slowly fill the disk. Reads also record more about how they went, so a problem reported after the fact can be traced instead of guessed at.
• The French-timbre voice is now labelled "French (Accent)" in the Voice menu. "French" previously appeared twice meaning two different things — a French-sounding English voice nested under English, and the actual French language listed separately.

Fixed

• Rostra starts in seconds again instead of stalling with an empty menu. When the system's audio was busy, opening the audio device could hold up startup for the better part of a minute — during which the menu bar showed "Engine: connecting…" and the Voice menu stayed empty. Rostra now starts without waiting on the audio device and attaches it on the first read.
• The app could start half-dead: empty Voice menu, gray icon, hotkey not responding. One internal fault at startup silently stopped every background-to-UI update in the app while playback kept working, which made it look like several unrelated bugs. Menus now reload their content instead of being carried across rebuilds, and a menu that ever comes up empty refills itself within a second.
• Rostra reads exactly what you highlighted in code editors and chat panels. In some apps a read could pick up far more than the selection — surrounding interface text and whole neighbouring messages — and speak that instead. The selection is now cross-checked before it is read, and anything that does not match falls back to a method that captures the highlight precisely.
• "Open at Login" now works on macOS. It was silently never taking effect — and the option was hidden, so there was no way to tell. Toggling it on now reliably adds Rostra to System Settings → Login Items (macOS asks once for permission to do so), and if anything stops that, Rostra opens the Login Items pane and says what to click instead of failing quietly.
• Rostra no longer freezes for half a minute and goes silent after audio devices change. Starting a read could stall roughly 31 seconds and then silently do nothing when the system's audio device was asleep or mid-switch (seen with and without Bluetooth headphones). The app now only re-opens the audio device when the device list has actually changed, caps how long a re-open may take, and shows "Audio device unavailable — try again" instead of failing silently.

0.1.1 — 13 August 2026

Added

• Rostra reads Bible scriptures the way a person would. Turn on Scripture Preset in Additional Settings — free tier — and verse numbers become landmarks instead of interruptions: announced when a verse begins a sentence, passed over silently when it lands mid-sentence, because a sentence in scripture routinely runs across several verses. References are spoken the way a person says them, and it works on text copied straight from the Bible sites people actually use. Off by default: the same "[1]" that marks a verse here marks a footnote in an academic paper.
• The Voice menu now tells you which voice is selected. It reads "Voice (Hazel — American English)" rather than just "Voice", so your current choice is visible at a glance instead of two submenus down.
• Speeds and rates are read as rates. "9.8 m/s" is spoken as metres per second, instead of reading the slash aloud. Prices like "$500/m" still mean per month.

Fixed

• Seven pronunciations reported from real use — durations, rates, abbreviations, and words whose correct reading depends on the sentence around them.

0.1.0 — 7 August 2026

Added

• Words that are spelled the same but said differently are now read correctly. "Play the record" and "Let me record it" no longer sound the same. Neither do content, read, close, wind, present, object, desert, progress, separate, permit, estimate and dozens more. Rostra now reads each one the way the sentence means it. Anything genuinely ambiguous keeps the reading it always had, so nothing that sounded right before sounds different now.

Improved

• Rostra is faster. The engine that turns text into speech has been rebuilt from the ground up. Reading starts sooner, long documents finish quicker, and the app is lighter on your machine.
• Windows: Support menu. Report a problem, share feedback, or copy diagnostics without leaving the app — the same options macOS already had.
• Windows: installing over a running copy now works. Upgrades no longer fail when Rostra is already open.

0.0.17 — 5 August 2026

Fixed

• Japanese voices now read kanji correctly. They used to mispronounce Japanese text with kanji in it — now they read it properly, with better pacing and punctuation in both Japanese and Chinese.

0.0.16 — 30 July 2026

Added

• Rostra now breathes like a person reading aloud. A long sentence with no punctuation gets a small breath before connecting words once it runs long enough that it would otherwise play as one unbroken sprint. Questions get a beat of air afterward so they land before the next sentence starts.
• Emphasis you can hear. Text marked bold or italic, and emphatically capitalized words, are spoken with a real push — a small pause plus stressed pronunciation.
• True high speeds up to 4× (Pro). A real 4×, not a dial that quietly stalls short of it — pauses keep their full length at every speed, so the text's structure stays audible even at a skim.
• Trading Preset (off by default). Signal and chart text reads the way a trader would say it — tickers, entry prices, timeframes, and stop/target shorthand all read naturally instead of colliding with dictionary words.
• A 55th voice — Timbre (French).

Fixed

• A batch of reading defects found in listening tests — dropped arrows in ranges, misread abbreviations, and a few date/number formats that read oddly.

0.0.15 — 27 July 2026

Fixed

• Rostra sometimes answered to your old shortcut as well as the new one after changing it. Fixed — changing your shortcut now always fully replaces the old one.

Changed

• Recording a custom shortcut is now a Pro feature. Free still includes a working set of preset shortcuts.

0.0.14 — 27 July 2026

Added

• Custom hotkey. Set your own Speak Selection shortcut from the menu, with live confirmation as you press it.
• Export format submenu. Export Selection as Audio now offers WAV and MP3 as separate menu items.

Fixed

• A rare issue where starting a new read while one was still playing could cut it off, repeat part of it, or go silent.
• The hotkey recorder capturing its own copy action while open.
• Selected text occasionally not being captured on the first press of the shortcut.