Orpheus Classic Driver

Orpheus Classic is an NVDA 2026 add-on for the older Dolphin Orpheus Build 18 speech runtime. It bundles the required Orpheus files and uses a dedicated 32-bit host process, allowing this old synthesizer to run from 64-bit NVDA.

The add-on appears as Orpheus Classic in NVDA's synthesizer list. It is separate from other Orpheus add-ons, so it can be installed and tested without replacing them.

Included runtime

The bundled runtime comes from an older SuperNova/Hal-era Orpheus tree. It is smaller and older than the later Orpheus 2.0x beta runtime.

Settings

Rate, volume, pitch, language, intonation, voicing, voice source, word pause, and phrase pause are exposed through NVDA's synthesizer settings. Audio is routed through NVDA's configured speech output device. Speech stop, pause, spelling, and Say All are supported through NVDA's normal commands.

Dictionary entries are stored outside nvda.ini. Each language has its own file, using Orpheus's language number:

%APPDATA%\nvda\orpheusClassic\dictionaries\00044.json

Open NVDA menu, Preferences, Settings, Orpheus Classic to select a dictionary language, edit its entries, import old Orpheus exception files, export dictionaries, save registry settings to the NVDA profile, or check for updates. During an upgrade, an older single dictionary.json is migrated once to UK English and then removed. If that file cannot be read, it is preserved with a .migrationFailed suffix and is not applied to speech.

Dictionary files

The dictionary editor supports JSON dictionaries for full Orpheus Classic compatibility. It can also import and export tab-separated Orpheus text dictionaries for easier sharing with older Orpheus setups. Automatic update checks can be set to Never, Hourly, or Daily. Daily is the default, and the manual Check for updates button remains available with every setting. Update information is cryptographically signed. NVDA also verifies package identity, version, file size, and SHA-256 hash before installing an update.

Notes

• This add-on is for NVDA 2026 and later.
• The classic runtime is old and may not respond as quickly as newer synthesizers.
• The add-on ships a baseline dictionary so common pronunciation fixes are available on a new install.
• The compatibility patches are limited to startup and add-on-contained runtime loading. They do not alter Orpheus voice data.

Changes

• 1.0.7:
  • Corrects the bundled UK English dictionary by removing inherited entries that made Bangor, Gaeilge, Jose, Kanye, latte, and lattes sound worse.
  • Corrects whole-word matching for arsehole and adds an AGI entry that reads the three letters separately.
  • Preserves existing user-edited pronunciations during the one-time dictionary update.
• 1.0.6:
  • Completes the move to separate language dictionaries by migrating and removing the obsolete shared dictionary.json.
  • Preserves a malformed old dictionary for recovery while preventing it from causing unwanted replacements.
• 1.0.5:
  • Adds separate dictionaries for each Orpheus language and preserves existing dictionaries as UK English during migration.
  • Removes an inherited correction that made Orpheus mispronounce beige.
  • Adds Never, Hourly, and Daily automatic update choices beside the manual update button.
  • Adds cryptographic verification of update information.
• 1.0.4:
  • Improves speech interruption after the Classic runtime has been idle, without requiring NVDA to restart.
  • Restores reliable spelling, punctuation handling, and Say All line advancement while preserving complete audio playback.
  • Preserves voice, rate, pitch, volume, intonation, voicing, and pause settings when the Classic host needs to recover.
  • Updates the bundled baseline dictionary with additional pronunciation corrections.
• 1.0.3:
  • Adds the Classic Orpheus voicing control to NVDA's synthesizer settings, with a cleaner default for new installs.
  • Adds the Classic Orpheus voice source control and limits it to the two values supported by the SAM driver.
  • Fixes long single-block speech stopping too early around heavy punctuation by waiting for the SAM runtime to report that speech is actually idle.
• 1.0.2:
  • Routes Classic Orpheus audio through NVDA's configured speech output device instead of the old runtime's default output path.
  • Fixes spelling and Say All completion by trusting the Classic host's final speech marker rather than falling back to estimated completion timing.
  • Improves spelling responsiveness by allowing single-character speech chunks to complete as soon as the Classic runtime has delivered its trailing audio.
  • Removes private host/proxy debug logging so normal use no longer writes orpheusClassic-audio.log in the Windows temp folder.
• 1.0.1:
  • Fixes the self-update helper so cleanup cannot delete the staged update folder before the external apply step replaces the running add-on.
• 1.0.0:
  • Initial public build of Orpheus Classic for NVDA 2026.
  • Bundles the old Orpheus Build 18 runtime and a dedicated 32-bit Classic host.
  • Adds NVDA settings for rate, volume, pitch, language, intonation, word pause, and phrase pause.
  • Adds an Orpheus Classic settings panel with dictionary editing, legacy exception import, dictionary import/export, update checks, and profile-backed registry settings.
  • Includes updated UK and US English language files from the later Orpheus language update package.
  • Adds timing guards so Say All is not allowed to race through long text immediately after speech starts.

Source code

The source archive for this add-on and its 32-bit host is available from onj.me/nvda/orpheusClassic-source.zip.