Microsoft seeks patent for AI that modifies live speech and synchronizes lip movements
Read more
Tecnoblog
tecnoblog.net

Microsoft seeks patent for AI that modifies live speech and synchronizes lip movements

Microsoft has filed a patent application regarding an artificial intelligence capable of altering the spoken content in real time during video broadcasts. This concept also includes the ability to adjust lip movements to match the newly generated speech.

The system, referred to by the service Patentlyze, would operate with a slight delay and has several potential applications. It could be used to translate speeches, eliminate verbal tics, and censor inappropriate terms, in addition to allowing the replacement of information presented during a presentation.

In the documentation, Microsoft characterizes this technology as a 'near real-time enhancement' mechanism intended for event presentations. The operation involves simultaneous analysis of both the audio and the images from the broadcast to identify segments requiring modification.

Upon detecting an expression that fits pre-configured rules, the AI would create an alternative version of the audio and adjust the mouth area to preserve synchronization. Although similar functionalities already exist in AI dubbing tools on platforms like Instagram, Microsoft's system promises to operate with a latency ranging between 1 and 10 seconds.

The patent claims cover everything from minor corrections, such as automatic removal of stutters or incorrect pronunciations, to filtering profanity and offensive language according to viewer preferences, as well as instant speech translation.

Additionally, the document describes a functionality that places the technology in the realm of deepfakes: the possibility of swapping information spoken by the speaker. According to Microsoft, this substitution would be based on data obtained through internet searches during the transmission itself.

This capability raises concerns about authenticity, as the system could produce multiple versions of a statement that was never made by the original presenter. Microsoft submitted the application on April 29 of last year, and the document was released by the United States Patent Office last week. However, according to Patentlyze, the claims were rejected in the initial preliminary review, although this does not end the process, as the company has the option to amend the application.

There is no indication that this technology is being developed for a specific product. As with other intellectual property registrations, the mere existence of the patent does not guarantee that the described system will be commercially launched.

Similar stories

Plaud Note Pro AI Device Review: Can It Replace a Smartphone?
Read more
www.aajtak.in

Plaud Note Pro AI Device Review: Can It Replace a Smartphone?

The question arises whether the Plaud Note Pro artificial intelligence gadget can replace a smartphone. If you frequently need to take notes during meetings, interviews, or important conversations, it is worth considering using a small AI device capable of recording the entire dialogue and then independently creating its transcription, summary, and to-do list.

This review examines the PLAUD Note Pro. This device functions as an AI Voice Recorder, specifically designed to record various types of communication, including business meetings, interviews, phone calls, and casual conversations.

The PLAUD Note Pro is equipped with four MEMS microphones, has 64 GB of internal storage, a 0.95-inch AMOLED display, and an AI transcription function. According to the company's statements, this device supports transcription in 112 languages.

New Artificial Intelligence Model Dessa Demonstrates Ability for Realistic Human Communication
Read more
iol.co.za

New Artificial Intelligence Model Dessa Demonstrates Ability for Realistic Human Communication

Existing artificial intelligence models excel at routine tasks such as document drafting, image generation, or coding assistance. However, a new product called Dessa has entered the market, offering a completely different level of human interaction.

Dessa is a conversational AI that sounded remarkably like a young woman during demonstrations. The intonation, word choice, and overall rhythm of speech seemed incredibly natural, not resembling a cold database containing the entirety of human knowledge.

When listening to the recording without knowing that an AI was involved, a listener could genuinely believe they were hearing a casual conversation between two real people. During a recorded dialogue with one of Dessa's developers, it even let out a spontaneous laugh mid-sentence, bypassing the standard requirement to wait for the interlocutor to finish their thought.

Unlike traditional voice assistants, which operate with predictable latency—they listen attentively, wait for complete silence, and respond in a clipped, strictly regulated manner—Dessa completely breaks this pattern. Interaction with it feels unprecedented and almost slightly unsettling precisely because of the subtle emotions and unrehearsed timing embedded in its delivery.

The model intentionally emphasizes the emotional nuances of human dialogue, actively listening and responding with genuine real-time presence. By mastering the organic rhythm and pace of conversation, Dessa does more than just improve voice synthesis; it sets a powerful new standard for how artificial intelligence will interact with us, offering a vivid glimpse into a future where humans and machines are virtually indistinguishable.

Popular