Scenario: Bringing Literature to Life with Immersive AI 

In many high schools across the United States, reading literary classics is a rite of passage—but for many students, it’s also a source of frustration. The dense, archaic language and complex themes can feel inaccessible, especially when delivered through static printed text or monotone classroom readings. As a result, students often miss the emotional depth, nuance, and relevance of the material. 

Now imagine that same classroom experience, reimagined through immersive AI. 

Instead of silently reading from a textbook, students log into an interactive platform powered by neural voice and avatar technology. Instantly, the characters of Biff from Death of a Salesman or El Sordo from For Whom the Bell Tolls come alive voiced with emotional realism, delivered in accents appropriate to the era, and animated with subtle expressions and gestures. Learners can control the pace of the performance, switch between voice styles (modern or classical), and pause to hear explanations or modern translations of challenging lines. 

Difficult passages are no longer stumbling blocks—they become conversation starters. Students can replay key scenes, compare character interpretations, and better understand the contextual situation. 

This isn’t just a more engaging way to learn literature—it’s a fundamental shift in how students connect with content. Through immersive tech, classics become more than required reading. It becomes a powerful, personal experience that deepens understanding, sparks curiosity, and brings timeless stories into the hearts and minds of a new generation. 

 

Architecture 

 

 

End-to-End Architecture Components for AI-Driven Avatar Video Creation 

This solution leverages multiple Azure services to securely process text or script inputs, generate high-quality speech, transform speech into animated avatars, and store the final video output. The architecture ensures data security, compliance, and scalability while enabling both low-code and developer-first automation. 

1. Identity & Access Management – Entra Integration

Azure Data Lake Storage (ADLS) integrates with Microsoft Entra for identity-based authentication and authorization. ADLS supports multiple mechanisms to control access: 

• • Shared Key Authorization – Uses the account access key for full control over the storage account. 

• • Shared Access Signature (SAS) – Provides time-bound, granular access to specific resources. 

• • Role-Based Access Control (Azure RBAC) – Assigns fine-grained permissions based on Azure roles. 

• • Attribute-Based Access Control (Azure ABAC) – Extends RBAC by applying conditional logic based on resource attributes. 

• • Access Control Lists (ACLs) – Allows more granular file/folder-level permissions within ADLS. 

This layered authorization ensures both least-privilege access and compliance with enterprise security policies. 

2. Threat Protection – Defender for Storage

Microsoft Defender for Storage scans uploaded files in real time to detect and prevent malicious content from entering the environment. This helps safeguard the storage layer from malware, viruses, and potential ransomware payloads. 

3. Central Data Repository – Azure Data Lake Storage

ADLS acts as the core data hub for storing all input and output artifacts: 

• • Input: .ssml (Speech Synthesis Markup Language) files, .txt files, or other supported formats. 

• • • .ssml: XML-based markup specifying pronunciation, pitch, rate, volume, and pauses to improve speech naturalness. 

• • • .txt: Plain text files containing unformatted script content. 

• • Output: Generated avatar videos in .mp4 format. 

Its hierarchical namespace and scalability make ADLS ideal for managing large multimedia workloads. 

4. Audio Content Creation – Azure Speech Studio

A low-code service for converting .ssml or .txt scripts into high-quality, lifelike speech. Features include: 

• • Multiple language support. 

• • Prebuilt and custom neural voices. 

• • Easy script editing and audio preview. 

Easy script editing and audio preview. 

For advanced automation, the Azure Speech SDK (developer-first approach) offers the same capabilities programmatically and can be integrated directly into CI/CD or data pipelines for end-to-end automation without manual intervention. 

5. Secure Network Access – Private Link

Azure Private Link ensures that all data flows between services occur over a private endpoint within the Azure network, preventing exposure to the public internet and reducing attack surfaces. 

6. Video Generation – Text-to-Speech Avatar Engine

Using Azure Speech Studio’s Avatar Engine, the generated audio is combined with animated lip movements, facial expressions, and gestures to create human-like avatar videos. 

• • Supports prebuilt avatars for quick deployment. 

• • Custom Avatar (optional, requires additional Azure resources): Mimics both the look and voice from training data for brand-specific personalities. 

• • Output is synchronized for natural, realistic engagement. 

7. Data Loss Prevention (DLP)

DLP policies restrict outbound connections from Azure AI Services to an allowlist of URLs, ensuring that sensitive or regulated content cannot be exfiltrated to unauthorized destinations. 

8. Output Rendering – MP4 Video Files

The final product is rendered in .mp4 format for maximum compatibility with distribution channels, including: 

• • Web browsers 

• • Social media platforms 

• • Video hosting services 

The .mp4 container supports synchronized video, audio, subtitles, and still images in a single file. 

9. Video Storage & Distribution

Once generated, .mp4 videos can be: 

• • Downloaded directly for immediate use. 

• • Archived back to ADLS for long-term storage. 

• • Stored in Cosmos DB or other Azure storage services for integration with content delivery systems or analytics pipelines. 

Workflow Description: Low-Code / No-Code Avatar Video Creation 

This workflow describes the end-to-end process for creating avatar-driven audio and video content using Azure services and SSML (Speech Synthesis Markup Language) files, culminating in downloadable video assets. The illustrated workflow follows a low-code / no-code approach, eliminating barriers for users without programming experience and enabling a wider audience to produce high-quality results. It is essential to follow Microsoft’s published guidance on Responsible AI for AI Speech’s Text-to-Speech and Avatar Services. This guidance covers critical considerations such as securing the solution’s network and access, carefully selecting and storing training data, and obtaining explicit consent from any voice actors whose voices are used. 

 

1. Input Preparation

• • Document Source: Textual content is prepared in either .ssml or .txt files. 
  • Storage: Files are uploaded to Azure Data Lake Storage for centralized access and processing. 

2. Speech Generation Using the AI Speech Studio

• • Audio Content Creation Tool: The. ssml or .txt files are manually uploaded to customize the speech to be spoken. The pitch, volume, speed, etc. can be adjusted to suit the needs of the user. The resulting preferences to the text will be saved to an .ssml file and exported.  

• • Avatar Service: The incoming .ssml file will be uploaded to the Avatar Service. Users can then select the avatar to speak the text. Hand gestures can also be added to emphasize points and create a more immersive experience. The final product is rendered as an .mp4 file that can be exported/downloaded for consumption. 

3. Output Management

• • Optional Storage: The video can be saved back to Azure Data Lake Storage for archival or further distribution. 

 

Security Considerations 

Azure Data Lake Storage: 

• Leverage Defender for Storage to monitor and identify unusual access, triggering security alerts for suspicious activity. 

• Use on-upload malware scanning to examine uploaded content and reduce the risk of malicious files entering the storage environment.  

• Enforce a minimum TLS version for secure communication with clients. 

• Use Role-Based Access Control (RBAC), Attribute based Access Control (ABAC) and Access Control Lists (ACLs) to manage permissions at various storage levels. 

• Use Entra security groups as the principal in ACL entries. This lets you manage users or service principals without reapplying ACLs to the directory. 

• It is recommended to enforce secure transfer for all storage accounts. With secure transfer enabled, all requests to the storage account are required to use HTTPS, and any HTTP-based requests will be denied. 

Azure AI Speech: 

• Configure private link for Azure AI Foundry to secure communications. 

• By default, Azure AI services resources accept connections from clients on any network. To limit access to selected networks, change the default action and configure VNET rules 

• Ensure that secrets and credentials are stored in secure locations such as Azure Key Vault, instead of embedding them into code or configuration files. 

• Leverage Azure AI services data loss prevention capabilities to configure the list of outbound URLs Azure AI services resources are allowed to access. This creates another level of control for customers to prevent data loss. 

Demo 

In this demo, a passage from For Whom the Bell Tolls is first converted into a .ssml file using the Audio Content Creation Tool in Azure Speech Studio (low-code / no-code approach). This .ssml file is then used to generate a Text-to-Speech Avatar, bringing the performance to life with synchronized lip movements, gestures, and expressions. 

https://youtu.be/ATVcRGHhyyw 

Related Use Cases  

• Product/Tool Onboarding: Educate users on how to use new tools and/or products. Interactive examples can be created to simulate real-world use cases. This can be leveraged in both corporate and consumer product spaces.  

• Corporate Memos: Important office updates can be sent out and delivered with avatars. The aim is to improve employee retention of the information.  

• Human Resources Departments: Rather than reading static, bland documents, employees can consume this same information with engaging avatars.  

 

Contributors 

This article is maintained by Microsoft. It was originally written by the following contributors.   

Principal authors:   

• Matt Kazanowsky | Cloud Solution Architect 

• Manasa Ramalinga | Senior Principal Cloud Solution Architect 

• Abed Sau | Principal Cloud Solution Architect 

• Oscar Shimabukuro | Senior Cloud Solution Architect 

• Anvita Kamat | Customer Success Account Manager 

• Susan Locke | Senior Account Executive