The US financial technology company Block has unveiled a new AI voice model named Dessa, distinguished by its ability to mimic human voice characteristics, including laughter, crying, and natural pauses during speech, as well as listening and speaking simultaneously, and operating directly on computers without the need for cloud server connections.
Local operation and synchronous interaction
According to a report published by the TechCrunch website today, Sunday, the model was developed by the research team at Block company. It was trained using human speech data, with the company stating that the model was designed to run locally on devices, including some MacBook models equipped with advanced processors. Dessa is characterized by its ability to listen to the user and speak with them simultaneously, enabling a smoother interaction than systems that rely on converting speech to text, then processing the text and producing a voice response.
Emotion and context simulation
The company believes this approach may help preserve aspects of human communication, such as tone of voice, rhythm, and emotional expression. Company researchers published demonstration recordings showing the model's ability to change its voice tone according to the conversation context, including mimicking exhaustion and emotional agitation, in an attempt to make interactions with voice assistants closer to natural human conversations.
Safety and privacy challenges
Running the model locally can reduce reliance on cloud servers, but the company has not announced its wide availability and has invited specialized safety researchers to test it, particularly regarding the risks of using artificial voices to deceive people or mimic human voices in ways that are difficult to distinguish. The development of such models reflects a trend toward making voice assistants more capable of understanding and responding to conversation contexts naturally, while challenges remain regarding safe usage, privacy, and distinguishing between real and AI-generated voices.
Source: Syrian Arab News Agency (SANA)
Comments (0)