Block's new AI model Dessa was trained entirely on human speech to catch emotions text-based models miss

Block, the fintech company behind Square and Cash App, has unveiled Dessa, a speech recognition model that ingests raw audio rather than relying on text transcription as an intermediate step. The model is designed to capture emotional nuances—such as sarcasm, hesitation, and pauses—that are typically lost when speech is flattened into text.

Most current speech-to-text pipelines discard paralinguistic cues during transcription, making it difficult for downstream systems to understand a speaker's true intent. By processing audio directly, Dessa aims to preserve these signals, which could improve customer service interactions, sentiment analysis, and any application where tone carries meaning.

The model reportedly runs locally on a MacBook Pro M4 Max, suggesting Block is exploring on-device inference rather than cloud-dependent processing. This approach could reduce latency and privacy concerns for real-time applications, though the source summary is truncated and details on performance benchmarks, training data scale, and deployment plans remain unclear.

Dessa represents a shift toward multimodal AI that goes beyond text-based language models, aligning with a broader industry trend of building systems that understand the full spectrum of human communication rather than relying solely on written or transcribed language.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.