
A new AI is monopolizing the attention of the digital world. It is about voice assistant developed by Sesamewhich stands out for its ability to Understand and adapt to real -time contextoffering a natural conversation with great expressive wealth.
Sesame is a technology company headed by Brendan Iribeco -founder and former executive director of Oculus VR; Ankit Kumarformer director of technology and co -founder of Ubiquity6; and Ryan Brownformer Director of Meta Reality Labs Research Engineering.
From Sesame, they claim that their philosophy argues that The advance of the conversational should be a collaborative effort. So, they have pledged to publish in open source His research, favoring experimentation and development. Its models will be available under an Apache 2.0 license.
At Sesame, We Believe in A Future Where Computers Are Lifelike. Today we are unveiling an early glympse of Our Expressive Voice Technology, Highlighting Our Focus on Lifelike Interactions and Our Vision for All-Day Wearable Voice Companions. https://t.co/edp8v8urgc pic.twitter.com/mc5nwnbjzm
– Sesame (@Sesame) Februry 27, 2025
In search of the perfect conversational
As explained in a statement published on its website, Sesame's goal is to “achieve the” voice presence “, the magical quality that makes spoken interactions feel real, understood and valued.” To do this, A conversational speech model are developed or CSM) able Not only to process applications, but also to participate in the conversation in a genuine way.
The four key features that Sesame contemplates to achieve this goal are: Emotional Intelligence (Understand user's mood), conversational dynamics (pauses, rhythm and emphasis), Contextual awareness (Adjust the tone according to the situation) and consistent personality (Maintain a recognizable style).
Likewise, the company recognizes that They have not yet created the model capable of meeting these requirements. Of course, they claim that they are on the right path to get it. «We are doing constant advances in several fronts, including Personality, memory, expressiveness and suitability. This demonstration is a sample of our work in conversational speech generation. The companions who are shown here have been optimized to be friendly and expressive in order to illustrate the potential of our approach ».
Maya and thousands, Sesame's voices
These “companions” of which Sesame speaks are the two voices that the company has developed for its voice assistant: Maya and thousands. Both are able to answer questions and maintain a fluid conversation using a natural tone and expressing expressiveness, thanks to the fact that more than 1 million audio hours have been used mostly in English for training.
At the moment, this conversational speech model It works in Englishalthough it can show a certain multilingual capacity due to the contamination of the data set with which it has trained. Sesame intends to Expand the compatibility of your assistant to more than 20 languages in the coming months.
In addition to this, they also intend to expand the size of the model, increase the volume of the data set and explore the ways of using previously trained language models, thus advancing towards large multimodal models that present a broad knowledge of both speech and text.
You can chat with both Maya and thousands through the Demo that Sesame has published on its website.
AI glasses to complete Sesame's experience
Another surprising novelty of Sesame is the fact that the company He is working on the creation of glasses that integrate the one that will complement their voice assistant. According to what is expressed by the technological, these glasses will be designed to «use throughout the day, providing high quality audio and convenient access to your partner that the world can see with you ».
It should be remembered that one of the people at the head of Sesame is Brendan Iribe, co -founder and former executive director of Oculus VR, a company that was acquired by goal and whose technology was used to develop the virtual reality glasses of the Mark Zuckerberg company.
Photo: GPT4