The French Start-up of Artificial Intelligence (AI), Mistral, launched its first models on Tuesday focused on vocal recognition and transcription in several languages.
This open source model, called Voxtral, allows you to transcribe audio, live or imported files in several languages that range from English to Hindi, automatically recognized.
You can also summarize, responding to the requests raised orally and Mistral, it intends to add other characteristics, such as the recognition of several interlocutors and their characteristics (age, sex) but also their emotions, according to a press release.
Improve commercial vocal systems
Voxtral can be used in particular to improve commercial vocal systems to respond to its customers on the phone, according to the new company. The French company also develops with the Stellantis car manufacturer a system that allows drivers to interact orally with an AI assistant in their vehicle.
The American Mastodon OpenAI presented for its part last year, a vocal mode for its GPT-4O model, capable of “reasoning” in real time through audio, vision and text. This version of Chatgpt can significantly read users’ emotions on the faces through the camera of a smartphone.
The French Research Laboratory in Artificial Intelligence Kyutai, founded by Xavier Niel, owner of the ILIAD group, and Rodolphe Saadé, CEO of the CMA CMA maritime transporter, presented in February a simultaneous translation model. Called “Hibiki” (“echo” in Japanese), this one translates the words of a real French user into English, as an interpreter would.
Source: BFM TV
