Cortana voice support

Cortana voice walks the user through the OOBE experience, enabling the user to complete parts of OOBE by responding to spoken prompts. Cortana voice during OOBE is currently available in the following languages: en-US, es-MX, ja-JP, en-GB, fr-FR, it-IT, de-DE, es-ES, fr-CA, en-CA, en-AU, pt-BR, zh-CN.

Note

With Windows 10 1903 and later, the Cortana voice-over will be disabled by default on Windows 10 Pro, Enterprise, and Education. The Cortana voice-over will still be enabled for Windows 10 Home editions.

The language value you set in OOBE.xml impacts the voice used during OOBE. The OOBE.xml value for language must be a language/region decimal ID associated with a Windows language pack. For example, the English (United States) language pack has an associated language/region decimal ID of 1033. For a full list of language/region decimal IDs that you can set in OOBE.xml, see Available Language Packs for Windows.

Cortana voice is enabled after the customer selects a language from the Language selection screen in OOBE. If the language selected by the customer, combined with the language in OOBE.xml, is supported by Cortana, Cortana will assist in that language upon entering the Region selection page.

Cortana voice will continue to assist throughout the OOBE process in that same supported language. Even if the user selects a region on the Region page that is not supported by Cortana, or selects a region that would cause Cortana to use a different accent after OOBE, Cortana voice will not change during OOBE.

If the language selected by the customer on the Language page combined with the language in OOBE.xml is not one of the supported combinations for Cortana, then the OOBE experience will be silent.

After the user completes OOBE, the voice used in the Cortana app will be based on the Language and Region selected during OOBE. At that point, Cortana will no longer consider the language in OOBE.xml.

Here are a few examples:

Language selected by customer (during OOBE) Language set in OOBE.xml Region selected by customer (during OOBE) Cortana voice assistance result (during OOBE) Cortana app voice result (after OOBE)
English 1033 (en-US language pack) 244 (US) en-US en-US
English 2057 (en-GB language pack) 244 (US) en-GB en-US
Russian 1049 (ru-RU language pack) 203 (RU) Silent Not supported

Disable Cortana voice support

For testing purposes, you can turn Cortana voice off, but you must enable it again before the device ships. To temporarily turn Cortana voice off, set the following registry key.

HKLM\SOFTWARE\Microsoft\Windows\CurrentVersion\OOBE : DisableVoice = 1 (DWORD)

Note

This setting should only be disabled for testing purposes. Shipping a device with Cortana voice support disabled is an unsupported configuration.

“Hey Cortana” feature

The “Hey Cortana” feature enables users to more easily engage Cortana on their Windows 10 device by speaking the phrase “Hey Cortana”.

For devices that meet hardware requirements, users have the option of enabling "Hey Cortana" during the OOBE flow, on the screen which asks the user if they’d like to make Cortana their personal assistant. The option is unchecked by default.

After OOBE, users can also enable "Hey Cortana" from Cortana & Search Settings. By default, "Hey Cortana" is not enabled.

Configure Hey Cortana

To optimize battery life, by default, Windows only asks users if they want to enable "Hey Cortana" on desktop devices with a microphone

For Windows 10, version 1709 and later, you an also include this option during OOBE if your device meets the policy requirement of including a hardware-offloaded key spotter (HW KWS).

For devices that meet this requirement, set the registry key:

HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Speech_OneCore\AudioPolicy : VoiceActivationIsBatteryCertified = 1.

To learn more, see Voice Activation.

Cortana voice recognition performance during OOBE

During the OOBE flow, Cortana may not recognize user speech as effectively as usual.

Cortana, like other Windows voice-enabled functionality, relies on automated speech recognition (ASR) systems that run both on the device and in the cloud. These systems include enhanced abilities to filter out background noise, recognize accented speech, and otherwise tell the difference between words and phrases that should be acted upon and words and phrases that should be ignored. However, to access the cloud ASR systems, Windows must have user consent and a connection to the internet. These may not be available during the OOBE flow. In that case, Windows uses a more limited, offline-only ASR system that does not have the level of robustness that Cortana usually offers.

While assessing input audio, an ASR system computes a numerical value known as a confidence score. Then it sets an acceptance threshold: Above the threshold, results are acted upon; Below the threshold, results are ignored. The more capable the ASR system, the better it can generally classify incoming data and assign well-differentiated confidence scores to results. Using this data, the ASR system improves its ability to act when it’s supposed to (a “correct accept”) and not act when it isn’t supposed to (a “correct reject”). The ASR system also uses the data to limit how often it acts when it shouldn’t (a “false accept”) or fails to act when it should (a “false reject”).

The less robust ASR system that the OOBE flow uses balances its false accepts and false rejects more aggressively. It’s very important for the OOBE flow to particularly minimize false accepts. The Windows 10 version 1809 release introduced a further adjustment to this balance that addressed reports of false accepts that occurred in noisy environments. This means that some utterances that the more robust, online Cortana and speech products would correctly accept end up falsely rejected by the more limited system. As a result, during the OOBE flow Cortana does not appear to correctly hear the user (especially in noisy environments).