Data, privacy, and security for Azure AI Voice Live API

Important

Non-English translations are provided for convenience only. Please consult the EN-US version of this document for the definitive version.

Note

This article is provided for informational purposes only and not for the purpose of providing legal advice. We strongly recommend seeking specialist legal advice when implementing Speech Services.

This article provides details regarding how data provided by you to the Azure AI Voice Live API ("Voice Live API") is processed, used, and stored.

Voice Live API is a fully managed service designed to empower developers to securely build, deploy, and scale high-quality, and extensible speech to speech experience for their voice agents. With Voice Live API, developers can choose from a list of different natively supported language models like GPT-Realtime, GPT-4.1, GPT-4o and GPT-4o-mini; incorporate an agent they have built using the Foundry Agent Service to give the agent speech-in and speech-out capabilities; or bring their own model of choice deployed in Microsoft Foundry.

Voice Live API stores and processes data to provide the service and to monitor for violations of the applicable Product Terms. See also the Microsoft Products and Services Data Protection Addendum, which governs data processing by the Foundry Tools, including Voice Live API. Voice Live API is an Azure service; learn more about applicable Azure compliance offerings.

Important

Voice Live supports connections to other Microsoft services and third-party services. Use of these services is subject to their respective terms and might result in data processing or storage outside of the Azure compliance boundary, as well as data flowing into the Azure compliance boundary.

The language models provided with Voice Live API are operated by Microsoft as an Azure service. If you choose to bring your own agent created with Agent Service or bring your deployed model in Foundry Models to Voice Live API, additional information on data, privacy, and security is available at Data, privacy, and security for Agent Service and Data, privacy, and security for use of models through the model catalog in Foundry.

What data does Azure AI Voice live API process?

Voice Live API processes the following types of data:

• Prompts and output. Prompts are submitted by the user, and content is generated by the GenAI model selected and converted to audio with or without avatar by Voice Live API.
• Uploaded data. You can provide your own data for use with Voice Live API using your own Azure Storage account or a configured data store, for example, your custom lexicon file to improve the pronunciation of the text to speech output, or your text and speech data to fine-tune the speech to text, text to speech and avatar model.
• External data. When you use the tools that support function calling, the service processes the outputs of those tools.

Important

Custom neural voice ("custom voice") and custom avatar are available with limited access. Learn more about data processing, storage and retention for custom text to speech (custom voice) and custom avatar.

How does Azure AI Voice live API process data?

The diagram below shows the data processing workflow for Voice Live API. It depicts how the API handles prompts (user audio input) through inferencing to produce content (agent audio output or video output with avatar), as well as how data from external tools is ingested into the service.

When these features are enabled by the user, Voice Live API processes audio input for noise suppression, echo cancellation, voice activity detection, and end of utterance detection , prior to sending the audio for speech recognition and language generation. For speech-to-speech models, audio output is generated directly from the language model. If a text-based language model is specified, Voice Live API converts the text response into audio. When an avatar is selected, the service streams the avatar and returns both the audio response and the avatar together.

When you bring your own model deployed in Foundry or an agent built with Agent Service to Voice Live API, the service interacts with the specified model endpoints to process your input prompts transcribed from audio and generate text output responses which may be further used or processed by Voice Live API for audio and avatar video generation. Data is processed for model inferencing in accordance with the terms that apply to the relevant model. Learn more at Data, privacy, and security for Azure OpenAI Service and Data, privacy, and security for use of models through the model catalog in Foundry portal.

To reduce the risk of harmful use of Voice Live API, the service includes content filtering support. The outputs processed by the service will be filtered in accordance with any content filtering that has been applied to the natively supported models, or the model deployment used by your Foundry Agent.

Data storage and retention

While Voice Live API itself does not store or retain customer data, the features (for example, custom voice, custom avatar, Foundry Agent) it interacts with may store customer data as the feature requires. Check data storage for custom voice, custom avatar, Foundry Agents, and Azure OpenAI if you are using these components. Learn more about locations of processing for ‘global’ and ‘data zone’ deployments.

Users can opt into a logging feature per debugging assistance from Microsoft engineers, when there is a support ticket filed. With this logging feature, users’ speech data is secured and stored in Azure storage managed by Microsoft within the same resource region. Microsoft’s debugging engineers are authorized Microsoft employees who access the data via point wise queries using request IDs, Secure Access Workstations (SAWs), and Just-In-Time (JIT) request approval granted by team managers. These logs are automatically removed in 30 days after generated.

To learn more about Microsoft's privacy and security commitments visit the Microsoft Trust Center.