AI news ·
Block Introduces Dessa Speech Model Trained Only on Human Conversation
Block built a speech model called Dessa using only real human conversation recordings, not synthetic audio, making its training data source a key product feature.
Block has introduced Dessa, a new speech model built entirely on recordings of human conversation. The model was not trained on synthetic audio, a departure from common practice in voice AI development.
The company’s decision to rely solely on human-sourced data highlights how the origin and makeup of training datasets are becoming product differentiators in the voice AI market. By emphasizing provenance, Block signals that dataset composition itself can serve as a competitive feature.