In a significant breakthrough for conversational AI, Block, formerly led by Jack Dorsey, has unveiled a new voice model named Dessa. Developed entirely in-house, this model operates locally, ensuring a high degree of authenticity with its ability to express emotions through laughter, pauses, and even tears.
Dessa is designed as a full-duplex model, enabling simultaneous listening and speaking, much like the recently launched GPT-Live model from OpenAI. Unlike many existing speech models that rely on a large language model for data retrieval, Block emphasizes that Dessa’s conversational ability is derived straight from its training data.
Current voice models often produce unnatural and irritating speech due to their reliance on a text-based pipeline for speech generation. Solutions like ElevenLabs v4 and Fish Audio Drama 3 typically use voice tags to infuse emotion into their outputs. In contrast, Block asserts that emotional depth is inherent within Dessa, although it is yet to be clarified whether this is managed through tagging.
“Interacting with current voice AI can just plain suck. It turns your voice into text and throws away an incredible amount of information text alone can’t carry. The most powerful AIs in the world can’t hear how you feel. We think that’s ridiculous, leads to a worse experience, and is potentially dangerous,” Block stated in a recent blog post.
Research teams at Block shared various interaction examples featuring Dessa, showcasing its ability to modify voice, cadence, and emotional expression. For instance, during one interaction, the model noted its fatigue from work, which was conveyed in its tone.
One sample showcased how the bot conversed about feeling comfortable, while its voice reflected a genuine emotional struggle. Another scenario involved a Block team member engaging in a game with the voice assistant.
Dessa is capable of running on standard consumer hardware, such as a MacBook M4 Max or comparable devices. Currently, public access to the model is limited as Block seeks input from safety researchers to investigate scenarios where a voice assistant might mislead users.
Block aims to leverage its innovative voice model to broaden its services within finance and enhance solutions for small businesses. With voice AI already facilitating order handling for Square, Dessa holds the potential to expand both business and consumer interactions more affordably.
Editor’s Take
This new development in voice AI from Block represents a significant leap forward in how machines understand and express human emotions. This could greatly enhance user experiences with virtual assistants, making interactions feel more natural and engaging. For businesses, integrating such technology could streamline operations and improve customer interactions, setting a new standard in the space of conversational AI.
Source: techcrunch.com
