# Dessa

> Full-duplex speech-to-speech AI model trained entirely on human conversation without text.

Dessa is a conversational AI model developed by Block that was trained from scratch strictly on human speech rather than text. Because it operates audio-in and audio-out in full-duplex mode, it can listen and speak simultaneously with realistic timing, backchanneling, and emotional tone. The model runs faster than real time and is compact enough to execute locally on high-end Apple silicon hardware.

Website: https://block.xyz/dessa

## What it is for

- Build natural voice-first conversational interfaces without text-to-speech latency
- Experiment with speech models that handle natural interruptions and emotional inflection
- Run real-time conversational voice agents locally on high-spec Apple hardware

## Features

- Full-duplex audio processing that listens and speaks simultaneously
- Trained strictly on raw sound without text transcripts or pretrained models
- Natural conversational dynamics including interruptions, laughing mid-sentence, and backchanneling
- Runs locally on a MacBook Pro with M4 Max or better
- Faster-than-real-time inference latency
- Trained on a single NVIDIA GB200 NVL72 setup

## Pricing

Free. Currently in research preview / waitlist; no pricing announced.

## Details

- Type: AI model
- Category: [AI](https://xiry.app/topics/ai.md)
- Open source: no
- Platforms: macOS, Web
- For: AI researchers, voice interface designers, and developers building conversational audio apps.
- Alternative to: OpenAI Advanced Voice Mode, Kyutai Moshi, Hume AI
- Tags: speech, voice, audio, full duplex, speech to speech, conversational ai

## Why it was saved

- [@jack, 23 October 2026](https://x.com/jack/status/2113697902916751661): Real, locally runnable frontier speech model for building voice-first products and natural conversational interfaces.

## Related

- [PersonaPlex-7B (nvidia/personaplex-7b-v1)](https://xiry.app/resources/personaplex-7b-nvidia-personaplex-7b-v1): Open full-duplex speech-to-speech conversational AI model by NVIDIA (AI model, Open source)
- [Adobe Podcast](https://xiry.app/resources/adobe-podcast): Web-based AI audio recording, editing, and speech enhancement tool (Web app, Freemium)
- [Airy](https://xiry.app/resources/airy): Fast, low-cost text-to-speech API and studio for creators and developers (AI model, Freemium)
- [cartesia-python](https://xiry.app/resources/cartesia-python): Official Python library for Cartesia's real-time text-to-speech API (GitHub repo, Paid)
- [ElevenLabs Conversational AI](https://xiry.app/resources/elevenlabs-conversational-ai): Platform for building and deploying interactive voice AI agents. (Web app, Freemium)
- [Hibiki-Zero](https://xiry.app/resources/hibiki-zero): Open-source real-time multilingual speech-to-speech translation model with voice preservation (AI model, Open source)

Updated 24 October 2026. Page: https://xiry.app/resources/dessa

Make your own library: Xiry takes your saved reels, tweets, Reddit bookmarks and Watch Later videos, makes them searchable and pulls the useful stuff out, like this library. It's invite-only for now; join the waitlist at https://xiry.app/#waitlist.