Block Unveils Tessa, First Speech-to-Speech AI Model Trained Without Text
Block has introduced Tessa, a real-time speech-to-speech AI model trained exclusively on human speech without text conversion, capable of detecting conversational nuances.
Block, the fintech company founded by Twitter creator Jack Dorsey, has introduced Tessa, an artificial intelligence voice model trained exclusively on human speech without using text. Block stated that Tessa is the world's first AI capable of a live, realistic conversation trained solely on human speech.
According to the company, Tessa can listen and respond in real time without first converting speech to text. This approach differs from conventional voice AI systems, which typically convert spoken words into text before processing a response. The model processes audio directly and can respond to conversational cues such as tone, pauses, laughter, and sarcasm.
Tessa utilizes a speech-to-speech architecture that allows it to listen and speak simultaneously. This design supports continuous exchanges between a human and the AI system, contrasting with separate steps of speech recognition, text processing, and speech generation.
Block noted that the model learned language from a significantly lower number of words compared to what large language models typically require. In February, Block announced the layoff of more than 4,000 employees, representing about half of its workforce, following the artificial intelligence revolution.
