NeuralVoices (Driver for Microsoft Neural Natural Voices)

NeuralVoices add-on is a driver for Microsoft's Neural Natural Voices. After installing the NeuralVoices driver, you need to install one or more neural natural voices. To use this type of speech synthesizer with others, you can use the MultiLang virtual driver.

MultiLang page in the Russian-language store

Donation

If you liked this improved version and want to express gratitude to Kvark, you can do so by transferring any amount to the following YooMoney wallet: https://yoomoney.ru/to/410012293543375

Version History:

2026.07.28

The Microsoft Neural Voices speech synthesizer driver has been updated to version 2026.07.28. Of the changes, only the libraries from the Speech SDK have been updated to version 1.51.1.

2026.07.02

The Microsoft Neural Voices driver has been updated to version 2026.07.02, which creates a separate instance of speechapi.SpeechSynthesizer for each voice add-on, even if several voices have the same license. It's pretty counterintuitive, though, what the hell is not joking. Here is a new version of the driver with this fix. I don't notice any difference at home, but maybe users who have experienced slowdowns before will notice an improvement in performance.

https://t.me/nvda_group/70936

2026.05.16

2026.03.24

2025.10.15

Publishing another update for the Microsoft Neural Voices synthesizer. This version fixes the engine's operation if the Microsoft Visual C++ Redistributable package is not installed on the system (the necessary DLLs are now included in the add-on). Also, the minimum required NVDA version in the manifest has been fixed (NVDA 2023.2 or higher is required).

https://t.me/nvda_group/60410

2025.10.14

The Microsoft Neural Voices synthesizer driver has been updated to version 2025.10.14. This build fixes the disappearance of some spoken phrases during a high number of simultaneous reading requests, observed, for example, during active text output in terminal emulator windows. https://t.me/nvda_group/60382

2025.10.04

The Microsoft Neural Voices driver has been updated to version 2025.10.04. The only change is compatibility with the latest 64-bit alpha builds of NVDA 2026.1.

https://t.me/nvda_group/59836

2025.08.30

Here is a new version of the driver for NVDA that can work with these voices.

Note: Required for the Dmitry voice, the updated Svetlana voice, and the English (US) voices Eric and Michelle. These voices require support for the new license. Support for previous licenses has been retained.

https://t.me/nvda_group/58355

2025.08.24

Updated the Microsoft Neural Voices driver to version 2025.08.24. The only change is a rollback of the Speech SDK to a more stable version, 1.44.

https://t.me/nvda_group/57994

2025.08.09

The Microsoft Neural Voices synthesizer driver has been updated to version 2025.08.09. Notable new features include:

  1. Microsoft Speech SDK updated to version 1.45.
  2. The driver now explicitly supports inserting pauses into speech by processing instances of the speech.commands.BreakCommand class.
  3. The synthesizer name in the synthesizer selection dialog (NVDA+Control+S) has been changed from "Neural Voices" to "Microsoft Neural Voices" for greater clarity.
  4. When selecting the default voice after the driver is loaded for the first time, the NVDA interface language is now taken into account. Where possible, a voice matching the NVDA language is selected.

https://t.me/nvda_group/56994

2025.07.05

Okay. If the current maximum volume is not sufficient, here is the next version where it has been significantly increased. Before updating, it is highly recommended to set the volume no higher than fifty percent in the speech settings.

https://t.me/nvda_group/55196

2025.07.04

Here is a version of the driver where I attempted to align the default rate, pitch, and volume values according to the internal representations of these settings. Hopefully, in MultiLang, they will no longer fluctuate from profile to profile. After updating, all these parameters will need to be adjusted again. Also fixed the behavior of the "read all" mode, where the cursor would jump far ahead of the text being read. This should no longer happen.

https://t.me/nvda_group/55163

2025.07.02

Yes, this key worked. Attached to this message is a new driver with Turkish, Russian, and three English voices. Everything should work.

https://t.me/nvda_group/54990

2025.06.10

When used in conjunction with MultiLang, it is better to use another driver for this synthesizer, where notifications about finishing reading text are generated more correctly. Also, the ability to install voices separately is a useful feature.

https://t.me/nvda_group/53216