Classic Voices 0.4.0

iOS 7 voices

• iOS 7.0's voices are here! I didn't add 7.1's voices because they are the same as the modern sounding voices.
• Get them with the iOS 7.0.6 firmware (iPhone 4) from Apple in the app, the NVDA add-on's data manager, Android, or classicvoices-data import ios70 on Linux; the iPhone, iPad, Mac and Vision Pro apps build them in.
• The iOS 7 enhanced voices change pitch themselves, so they don't hand pitched words to the compact voice as the iOS 5 and 6 enhanced voices do.

Speech commands without pauses

• Screen reader indexes (link sounds from Unspoken, NVDA's say-all, Orca's and SAPI's bookmarks, iOS and Mac bookmarks) no longer break a sentence in two: the sentence is spoken as one, and each index is reported as the audio reaches it.
• Rate, volume and pitch changes happen in the middle of a sentence on the iOS 5 and later voices and the Mac OS X voices. The iOS 3 and 4 engines can't do that, so there the sentence is still split, with at most 100 ms between the parts.
• More of NVDA's speech commands: character mode (spelling), phonemes (MathCAT's "x to the 3rd"), breaks, rate, volume and pitch. SAPI 5 and Speech Dispatcher get the same through their own markup: bookmarks, silences, prosody, spelling and phonemes.
• MacinTalk voices honour rate, volume, spelling and phonemes through MacinTalk's own [[ ]] commands.
• Phonemes now work with the iOS 4 voices too.

Pronunciation rules

• Each voice uses the pronunciation rules its own release had: the iOS 3 voices iOS 3's, the iOS 4 voices iOS 4's, the iOS 5 and 6 voices theirs, and each Mac OS X voice its own. Before, every voice used iOS 6's.
• New setting, "Newer pronunciation rules", off by default: Apple's later pronunciation dictionary and rules, added to any voice's own, the iOS 3 and 4 voices included. Turning it on downloads them from Apple for every language (about 5 MB). It adds no noticeable delay: the rules are checked against the text before any of them runs, so a phrase costs well under a millisecond.
• On iPhone, iPad, Mac and Vision Pro both rule switches are also on each voice's page.

Other changes

• The NVDA add-on, the Windows and Linux app and the Android app say when a newer Classic Voices is out, once per version, and offer to open its download page. They check GitHub at most once a day and never download or install anything themselves.
• iPhone, iPad and Mac app: Speak and "Play sample" work for every downloaded voice again. Voices downloaded since the speech extension started keeping them were silent in the app, though they spoke everywhere else.
• Windows and Linux app: choosing a version in the Version list no longer throws you out of the list (it said "panel" instead of the version).
• Windows and Linux app: a download's progress bar no longer goes quiet when you open another language: every page ends with an "In progress" section with a progress bar for each running download.
• The engines use far less memory when loading large dictionaries and rule sets.
• Windows and Linux app: errors and crashes are written to a log that Diagnostics shows.

Downloads

• classicvoices.nvda-addon: NVDA 2023.1 or later.
• ClassicVoicesSetup.exe: the Windows app and the SAPI 5 voices.
• ClassicVoices.apk: Android 5 or later.
• classicvoices-linux-x86_64.tar.gz and classicvoices-linux-aarch64.tar.gz: Speech Dispatcher module and app.
• The iPhone, iPad, Mac and Vision Pro apps are built from this repository on GitHub (see the README).

Classic Voices 0.3.0 (testing)

A test release for Windows, NVDA, Android and Linux. Nothing of Apple's is in these files or in the repository: each app gets the speech engines and voices on your device, from Apple's own servers or from a firmware file you have.

What's new

• Fast engine code everywhere, built for you. After the firmware is in, every app makes the fast engine code by itself on your own device: the Windows app, the Linux app, the NVDA add-on and the Android app. On Android it's built on the phone. The voices keep working while it's built (a few minutes), and the fast code takes over at the next thing spoken. It speaks the same as before, only faster: on an Android phone, iOS 6 started speaking in a tenth of a second instead of a second and a half.
• The NVDA add-on gets the firmware from Apple itself (Tools, Classic Voices data, Get firmware from Apple) and has a Build fast code button. You can close its dialog while fast code is built; NVDA says when it's ready.
• iPhone, iPad and Mac builds (your own build on GitHub) now include Tiger's, Leopard's and Snow Leopard's MacinTalk voices, read from the install discs kept on archive.org, and their fast engine code (the build on GitHub now takes about an hour and a quarter). Alex for Leopard, Snow Leopard and Lion is a download in the app's macOS tab, in the background like any other voice.
• iPhone: downloaded voices kept working after a restart only once the app was opened. Fixed.

Files

• ClassicVoicesSetup.exe, for Windows (tested on Windows 11). It installs the Classic Voices app in the Start menu, and the voices for Windows programs (SAPI 5, for 64- and 32-bit programs), so JAWS, NVDA's SAPI option and other speech programs can use them. The installer isn't code-signed yet, so Windows may say the publisher is unknown: choose More info, then Run anyway.
• classicvoices-0.3.0.nvda-addon, the NVDA add-on (NVDA 2023.1 or later). Open it with NVDA running. Then go to the NVDA menu, Tools, Classic Voices data. If the Windows app is installed too, the add-on uses the app's voices.
• ClassicVoices.apk, for Android 5 or later (fast code on 64-bit phones). Allow your browser or files app to install apps. Then open Classic Voices, get the firmware, pick voices, and choose Classic Voices as the text-to-speech engine.
• classicvoices-linux-x86_64.tar.gz (most PCs) and classicvoices-linux-aarch64.tar.gz (64-bit ARM, such as a Raspberry Pi 4 or 5), for Linux with Speech Dispatcher 0.11 or later (Debian 12, Ubuntu 22.04, Fedora 36 or newer). Unpack it and run classicvoices/install.sh: it installs into your home folder, no root needed. Then open Classic Voices from your applications menu (it needs wxPython: sudo apt install python3-wxgtk4.0), or run classicvoices-data import ios6 in a terminal, and choose Classic Voices as Orca's speech synthesizer. Tested with Orca on Ubuntu 24.04. More in linux/README.md.
• iPhone, iPad, Mac and Vision Pro: those builds contain Apple's code, so they're never published here. Build your own with the one-line script in the README.

Classic Voices 0.2.0 (testing)

A test release for Windows, NVDA and Android. Nothing of Apple's is in these files: each app gets the speech engines and voices from Apple's own servers (or from a firmware file you have) on your device.

Files

• ClassicVoicesSetup.exe, for Windows (tested on Windows 11). It installs the Classic Voices app in the Start menu, and the voices for Windows programs (SAPI 5, for 64- and 32-bit programs), so JAWS, NVDA's SAPI option and other speech programs can use them. The installer isn't code-signed yet, so Windows may say the publisher is unknown: choose More info, then Run anyway.
• classicvoices-0.2.0.nvda-addon, the NVDA add-on (NVDA 2023.1 or later). Open it with NVDA running. Then go to the NVDA menu, Tools, Classic Voices data. If the Windows app is installed too, the add-on uses the app's voices.
• ClassicVoices.apk, for Android 5 or later. Allow your browser or files app to install apps. Then open Classic Voices, get the firmware, pick voices, and choose Classic Voices as the text-to-speech engine.
• Linux: install from the source, see linux/README.md.
• iPhone, iPad, Mac and Vision Pro: those builds contain Apple's code, so they're never published here. Build your own with the one-line script in the README.

Every app has the same layout: an iOS tab, a macOS tab and Settings, then a version, a language, and each voice.

Fast engine code isn't built for you automatically yet. On Windows and Linux, use Speed up the voices in the app.