Block Unveils Dessa: An AI Voice Model That Expresses Emotion
A New Chapter in Voice Technology
Jack Dorsey’s company, Block, has introduced a new voice-based artificial intelligence called Dessa, which marks a significant departure from current industry standards. The model is distinguished by its ability to naturally express emotions during conversation: it can laugh, incorporate pauses, and even speak with the intonation of someone crying. According to Block, the model was trained entirely on human speech data and requires only a single NVIDIA GPU to function.
Technical Advantages and Full-Duplex Capabilities
Dessa is a "full-duplex" system, meaning it can listen and speak simultaneously. While this capability is similar to OpenAI’s GPT-Live, the underlying approach differs. Traditional voice models often rely on text-to-speech conversion, which can make interactions sound robotic and unnatural. The team at Block argues that it is impossible to convey emotion through text alone; therefore, Dessa learned to express emotions directly from speech data without the need for additional tags.
In demo recordings released by the company, the model speaks with varying timbres and moods. For instance, when the bot mentions being tired, its voice sounds genuinely weary. In other instances, the model speaks with such emotional depth that it mimics the sound of someone on the verge of tears, creating a highly realistic impression for the listener.
The Advantage of Local Processing
One of the most significant aspects of Dessa is that it can run on standard consumer hardware, such as a MacBook M4 Max. This means data does not need to be constantly sent to cloud servers, which enhances privacy and reduces latency.
At present, the model is not publicly available. Block has invited security specialists to participate in testing to mitigate risks, such as the potential for the voice agent to be used to deceive individuals.
What This Means for Business
Block, which operates in the financial technology sector and owns Square, already utilizes voice technology for order management. Developing its own locally-run model will allow the company to significantly reduce costs and offer businesses and consumers higher-quality, personalized services. Although the technology is still in its early stages, it clearly points to a trend where artificial intelligence is acquiring increasingly "human" traits in its interactions with people.