Infovox 230 → NVDA 2026.1 add-on

Brings the long-abandoned Infovox 230 text-to-speech engine (Telia Promotor / Babel-Infovox, both defunct; the engine is public domain) back to life as a self-contained synthesizer add-on for NVDA 2026.1 — 60 voices across 12 languages, no hardware dongle, no CrypKey, no 16-bit license manager, and no administrator rights required.

Languages: American & British English, German, French, Italian, Castilian Spanish, Dutch, Danish, Norwegian, Swedish, Finnish, Icelandic — each with five formant voices (Male, Female, Giant, Child, Zombie).

Install

In NVDA: NVDA menu → Tools → Add-on Store → Install from external source, and choose infovox230-0.1.nvda-addon.
Restart NVDA when prompted.
NVDA menu → Preferences → Settings → Speech, set Synthesizer to Infovox 230 (Telia Promotor), then pick a voice and adjust rate / pitch / volume.

The first time you select the synthesizer it builds its voice table and starts the engine — this takes a few seconds once; afterwards it is immediate. No UAC prompt appears.

Nothing is written to the Windows registry. The engine insists on reading its configuration from HKCU\Software\Babel-Infovox AB\Infovox 230, so the host loads a private hive file (infovox230.hive, inside the add-on folder) with RegLoadAppKey and points its own process's HKEY_CURRENT_USER at it with RegOverridePredefKey. The engine's registry reads land in that file; your real registry is never read or written, and the redirection disappears when the host process exits. Removing the add-on folder removes every trace. Keys written by earlier versions of this add-on are deleted automatically on first run.

How it works

NVDA 2026.1 is 64-bit and cannot load the 32-bit Infovox SAPI4 engine in-process — which is exactly why every SAPI4 voice went silent in 2026.1. The add-on bridges that gap:

 NVDA 2026.1 (64-bit)                          bundled host (32-bit Python)
 ┌───────────────────────────┐   127.0.0.1 TCP  ┌────────────────────────────┐
 │ synthDrivers/infovox230.py │ ───────────────► │ host/infovox_host.py       │
 │  • lists 60 voices         │  control frames  │  • voice table→private hive│
 │  • builds tagged text      │                  │  • LoadLibrary Ivx230nt.dll│
 │  • nvwave.WavePlayer        │ ◄─────────────── │  • ITTSCentral.TextData     │
 │  • fires index / done       │  PCM + marks     │  • captures PCM (IAudioDest)│
 └───────────────────────────┘                  └────────────────────────────┘

Everything is bundled inside the add-on: the engine and its 12 rule sets, a 32-bit Python with comtypes, the host, the 64-bit driver, and a tiny emulated Sx32w.dll. See ARCHITECTURE.md for the full reverse-engineering write-up.

Licensing defeated + engine fixes (all in `Ivx230nt.dll`)

The engine ran on Windows 95/98/XP only and was gated by a Rainbow Sentinel SuperPro hardware dongle + CrypKey, with a 16-bit license manager that will not run on Windows 11. Four minimal, surgical binary patches (documented in ARCHITECTURE.md, and reproducible from the originals) make it run free and self-contained on Windows 11:

| Offset | What | Fix | |--------|------|-----| | 0x15b50 | Rainbow SuperPro dongle check | always report authorized | | 0x0a4c9 | GetUlexInstance treats missing user lexicon as fatal | make it non-fatal (lexicon is optional) | | 0x09dec | Init rejects mode unless licence-type 1/2/3 | accept and proceed | | 10 sites | engine reads config/voices/lexicon from HKLM | read per-user HKCU (no admin), which the host then redirects into a private hive file so the real registry is untouched |

The voice definitions themselves come from the engine's own trial installer, which does run on Windows 11 and supplied the correct speaker/formant data.

Folder contents

nvda-infovox230/
  infovox230-0.1.nvda-addon      the installable add-on  ← install this
  README.md                      this file
  ARCHITECTURE.md                how it works + reverse-engineering notes
  addon/                         the add-on source tree (engine, host, driver, python32, modes.json)
  host/                          host + SAPI4 interface source (mirrors addon/…/host)
  sx32w_stub/                    the dongle emulator + its generator
  tools/                         diagnostics + the test harnesses used to bring it up
  _run/                          proof WAVs and logs from bring-up (first_speech.wav, all_voices.wav, …)

Status