Block’s new AI model Dessa was trained entirely on human speech to catch emotions text-based models miss
What it does
Block, known for Square and Cash App, has developed Dessa, an AI model trained exclusively on human speech instead of text transcripts. Unlike typical speech-to-text models, Dessa processes audio input directly to detect emotional cues such as sarcasm, pauses, and tone in real time. The model runs locally on a MacBook Pro M4 Max, bypassing the need for cloud processing. It is not yet available to the public but signals a new approach to audio-based AI interpretation.
Why it matters
Most voice AI systems convert speech to text and then analyze the text, losing subtle vocal information that conveys emotion. Dessa’s direct audio analysis captures these nuances, potentially improving applications like customer support, virtual assistants, or any system that relies on understanding user sentiment. Running locally also reduces latency and data privacy risks, which is crucial for sensitive environments. This approach raises the bar on how AI can interpret human communication beyond words alone.
Who it is for
Dessa’s design suits builders and businesses needing richer voice analysis to improve user experience or operational insights. Companies handling customer calls, interview analysis, or mental health monitoring could gain by integrating models that understand emotion, not just semantics. The local execution fit also benefits companies cautious about cloud data exposure or high bandwidth demands. Early adopters in voice AI and human-machine interaction fields should track this development closely.
The catch
Dessa is still in development and not yet publicly accessible, limiting immediate integration. Its performance on devices outside the MacBook Pro M4 Max or at scale is unproven. Training solely on human speech emphasizes nuances but may restrict generalization for other types of audio input. The model’s complexity could challenge smaller teams lacking access to powerful hardware or speech data sets. Commercial impact will depend on how quickly Block opens access and how well Dessa performs in real-world conditions.
What to watch next
Watch for announcements on public availability and detailed performance metrics. If Dessa delivers on real-time, emotion-aware speech interpretation at scale, it will pressure competitors relying on text-first models to innovate. Monitor developments around on-device AI for privacy and latency advantages. Also keep an eye on potential partnerships or integrations with communication platforms and customer service technologies, where emotional nuance adds clear value.
AI Quick Briefs Editorial Desk