Microsoft Debuts Advanced Deepfake Generation Tool at Ignite 2023 Event

Microsoft Ignite 2023 has unveiled a surprising innovation—a tool capable of crafting photorealistic avatars that can be animated to say things they never actually uttered.

Google Chrome to Transition from Manifest V2 to Manifest V3 by 2024

The tool, named Azure AI Speech text-to-speech avatar, is now available in public preview. Users can generate videos wherein avatars speak by uploading images to mimic someone's appearance and providing a script. What powers the animation is a specially trained model, while a dedicated text-to-speech engine audibly delivers the script.

According to an announcement, "With text to speech avatar, users can more efficiently create videos... to build training videos, product introductions, customer testimonials simply with text input." Microsoft explains further in a blog post why avatars are also ideal for creating virtual assistants and chatbots.

Samsung Set to Unveil Galaxy S24 Series: Pre-Order Dates Revealed

The avatars come with multilingual capabilities. They can leverage AI models like OpenAI’s GPT-3.5 to field unscripted queries when used in chatbot implementations.

Ethical Implications

The unveiling of this feature has sparked a wave of ethical concerns.

Affidea Announces Creation of 120 New Jobs in Belfast Expansion

One pressing issue is the use of AI to create digital likenesses, reminiscent of the disputes in the recent SAG-AFTRA strike. Actors now receive payment for AI-generated likenesses under new agreements with studios. However, questions linger about Microsoft's stance on the use of actors' likenesses by its customers.

When questioned about the policy for customers using actors' likenesses, perhaps without deemed adequate compensation or notification, Microsoft remained silent. The tech giant did not specify whether there would be a requirement for these avatars to be labeled as AI-generated, a practice adopted by platforms such as YouTube and others.

Following up, Microsoft shared that customers creating custom avatars must have "explicit written permission" from the subjects and consent for the usage scope. There is also a mandate for disclosing the AI-generated nature of the avatars.

Launching the "Personal Voice" Capability

A similarly guarded approach is applied to the Personal Voice feature, another generative AI tool also making its debut at Ignite 2023.

In-Depth Analysis: The Moto G54 Power Edition Performance and Features

Personal Voice allows users to clone their voice using just a one-minute sample. Intended uses include creating voice assistants, content dubbing in multiple languages, and unique narration for storytelling. To prevent misuse, prerecorded speeches are not permitted, users must provide explicit consent, and Microsoft will verify the consent matches the training data before synthesis is allowed.

Microsoft imposes restrictions on the feature's usage, stating in a blog post, "Voice model usage must remain within an application and output must not be publishable or shareable from the application." Customers must comply with certain criteria to have control over the creation and use of voice models.

Initially, the company didn’t respond to inquiries regarding actor compensation for voice contributions. Later updates revealed intentions to add watermarks to personal voices to aid in identification. However, the integration of watermark detection into applications and platforms is contingent upon Microsoft's approval—a potential limitation for adopters of the service.