GitLab Transcend - Intelligent orchestration
Mini toolsRaon-Speech
By KRAFTON
Raon-Speech is KRAFTON’s 9B bilingual SpeechLM designed for both speech understanding and speech generation. The repository says it supports STT, TTS, SpeechChat, and TextQA in one model family, and was trained on more than 1 million hours of curated English-Korean speech-text data. It uses a staged LLM-to-SpeechLM training recipe and is positioned as state of the art on average across 42 speech and text benchmarks against similarly sized baselines. The repo also highlights low-latency streaming TTS, faster-than-real-time synthesis on single-GPU setups, and open release of checkpoints, code, demo tooling, and Korean speech benchmarks.
Multimodal
Gen 3
Overview
Raon-Speech is KRAFTON’s open-source 9B bilingual speech language model for English and Korean. It is built as a unified speech AI system that supports speech-to-text, text-to-speech, spoken QA, speech chat, and text QA, with low-latency generation and optional speaker-conditioned TTS.
About KRAFTON
View Company ProfileTools using Raon-Speech
No tools found for this model yet.
Last updated: April 8, 2026
