@DanKornas: Writing a two-person audio demo shouldn’t mean stitching together separate speech clips. Dia is a 1.6B-parameter text-t…

X AI KOLs Timeline Models
text-to-speech two-speaker open-source dialogue-generation audio-model research

Summary

Dia is a 1.6B-parameter open-source text-to-speech model that generates English dialogue from transcripts, supporting two-speaker generation, audio conditioning, and nonverbal cues.

Writing a two-person audio demo shouldn’t mean stitching together separate speech clips. Dia is a 1.6B-parameter text-to-speech model for researchers and builders generating English dialogue from transcripts. It helps you create scripted conversations by alternating [S1] and [S2] tags, with optional audio prompts for voice, emotion, and tone conditioning. Key features: • Two-speaker generation – turns tagged transcripts into a dialogue track • Audio conditioning – accepts a 5–10-second prompt plus its transcript • Nonverbal cues – supports cues such as laughs, coughs, sighs, and applause • Flexible interfaces – examples cover Transformers, CLI, and a Gradio UI • Public artifacts – includes inference code and links to pretrained checkpoints The repo is open-source (Apache 2.0 license). The README says the model currently supports English and has only been tested on GPUs; it also prohibits identity misuse, deceptive content, and illegal or malicious use. Link in the reply
Original Article
View Cached Full Text

Cached at: 07/22/26, 04:34 PM

Writing a two-person audio demo shouldn’t mean stitching together separate speech clips.

Dia is a 1.6B-parameter text-to-speech model for researchers and builders generating English dialogue from transcripts.

It helps you create scripted conversations by alternating [S1] and [S2] tags, with optional audio prompts for voice, emotion, and tone conditioning.

Key features: • Two-speaker generation – turns tagged transcripts into a dialogue track • Audio conditioning – accepts a 5–10-second prompt plus its transcript • Nonverbal cues – supports cues such as laughs, coughs, sighs, and applause • Flexible interfaces – examples cover Transformers, CLI, and a Gradio UI • Public artifacts – includes inference code and links to pretrained checkpoints

The repo is open-source (Apache 2.0 license). The README says the model currently supports English and has only been tested on GPUs; it also prohibits identity misuse, deceptive content, and illegal or malicious use.

Link in the reply

GitHub: https://github.com/nari-labs/dia

If you’re into AI, ML, agents, and building real systems, join my newsletter (it’s free): http://dankornas.substack.com

Similar Articles

SpeechDx: A Multi-Task Benchmark for Clinical Speech AI

arXiv cs.AI

SpeechDx is a large-scale benchmark for clinical speech AI spanning 12 datasets and 27 tasks across diverse health conditions, structured by stages of speech production. It evaluates 12 state-of-the-art audio encoders and shows that current models do not generalize reliably across the clinical speech landscape.