# Speech-First Voice Models Block Trains Dessa for Direct Audio Replies By Amy Duong | Written with AI assistance | Published 2026-10-26 | Tech Source: Trend Hunter, https://www.trendhunter.com/trends/speech-first-voice-models References: [gizmodo](https://gizmodo.com/check-out-the-many-uncanny-human-ish-voices-from-blocks-new-ai-model-2000830237)  Block's Dessa handles spoken input as audio instead of turning it into text first, letting people talk over it and keep a more natural back-and-forth. The voice model was built from human conversation rather than text-first data, giving its replies more feeling, pauses and everyday cadence. It also uses duplex speech architecture, meaning it can listen and answer at the same time instead of dropping into silence when interrupted. Initial access sits with a handful of AI safety researchers, and the model runs locally on MacBook Pro systems using M4 Max or higher hardware. Local support on MacBook Pro systems turns Dessa into a voice-safety test bed that does not need cloud infrastructure. Duplex speech also shifts the category beyond text-based call and response, making interruption handling and vocal nuance part of what teams evaluate in voice AI. Image Credit: Block ## Trend Insights (Trend Hunter) - Score: 4.7/10 - Popularity: 25% | Activity: 17% | Freshness: 100% - Audience gender: 50% men, 50% women - Primary generations: Millennial, Gen X - Top markets: North America ## Categories [Trend Hunter](https://www.trendhunter.com/trends) > [Tech](https://www.trendhunter.com/tech) > [AI](https://www.trendhunter.com/ai) ## Key Themes ### What's Driving This Trend - **Speech-native AI:** Voice systems trained directly on human audio create openings for more emotionally aware assistants that preserve cadence, timing and conversational nuance. - **Duplex Voice Interaction:** Simultaneous listening and speaking enables more fluid exchanges where interruption handling becomes a key differentiator for customer service, education and accessibility tools. - **Local Voice Safety Testing:** On-device deployment gives researchers and enterprises a lower-friction path to evaluate sensitive voice AI behavior without relying on cloud infrastructure. ### Who This Affects Most - **Artificial Intelligence:** Speech-first model design expands the competitive landscape beyond text-based LLMs by making vocal realism and real-time dialogue core product capabilities. - **Consumer Electronics:** High-performance laptops and edge devices gain new relevance as local hardware becomes capable of running advanced conversational voice models. - **Customer Experience:** Natural turn-taking and interruption-aware voice agents point to more human-like support systems that can reduce friction in live service interactions. ## Related on Trend Hunter - [Emoji-Like AI Assistants](https://www.trendhunter.com/trends/avatar-ai-assistants.md) - [AI Voice Generators](https://www.trendhunter.com/trends/arcana.md) - [AI Voice Platforms](https://www.trendhunter.com/trends/all-voice-lab.md) - [MAI Voice and Vision Models](https://www.trendhunter.com/trends/mai-transcribe.md) - [Upgraded AI Voice Models](https://www.trendhunter.com/trends/gptlive1.md) - [Always-On AI Assistants](https://www.trendhunter.com/trends/agentic-ai-assistants.md) - [AI Routing Assistants](https://www.trendhunter.com/trends/mezzie.md) - [Multi-Model AI Discussion Platforms](https://www.trendhunter.com/trends/discursa.md) - [Advanced AI Models](https://www.trendhunter.com/trends/deephermes-3.md)