← Back to Voice models
Krafton
Raon-SpeechChat 9B
Release date unpublished
KRAFTON's open-weight full-duplex speech language model for real-time English conversation, with natural turn-taking, backchannels and barge-in. It's built on the Raon-Speech base model.
- Price per minute
- —
- Pricing
- Open weights; no hosted price.
- Inputs
- audio
- Tool calling
- —
- Languages
- 1
- Voices
- —
- Big Bench Audio
- 57.5%
- Access
- Research only
- Licence
- Creative Commons Attribution-NonCommercial 4.0
Strengths & weaknesses
Strengths
- Open checkpoints, training and inference pipeline
- 17 built-in personas plus voice cloning from reference audio
- Strongest on Full-Duplex-Bench v1.0 turn-taking and interruption tasks (KRAFTON report)
Weaknesses
- Non-commercial licence (CC BY-NC 4.0)
- English only
- Big Bench Audio 57.5% (AA)
Not notable: #37 on AA Big Bench Audio (57.5%). releaseDate is null: the HF card cites 'April 2026' with a technical report 'Published Apr 8', while the arXiv ID 2605.23912 implies May 2026. Training data: 119K hours (report) vs 116K hours (GitHub README). About 10B parameters despite the 9B name. No tool calling is mentioned, so toolCalling is null.
Links
API model strings
- huggingface
KRAFTON/Raon-SpeechChat-9B