LocalClaw/TTS/A.X K2 Raon-Speech 21B-A3B
Local ASR model

A.X K2 Raon-Speech 21B-A3B

Catalogue summary: Bilingual English/Korean speech-language model for local STT, TTS, SpeechQA, SpokenQA and turn-based multimodal chat. The MoE checkpoint has 21.2B total parameters with about 3.5B active, includes KRAFTON's AuT speech encoder and Mimi-style codec, and loads through official Transformers remote-code pipeline examples.

Repository editorial metadata; verify comparative claims in the linked upstream material.

GPU recommendedspeech-to-text transcription2 languagesCC-BY-NC-4.0
Choose an app

Start with the recommended app, or choose another supported option.

Compare speech models
Open model filesRecommended verified starting pointOpen in UnslothLaunches Unsloth Desktop on this modelOpen official projectApplication source and guided setupOpen on Hugging FaceFiles, licence and model cardCheck my machineOptional: verify the fit with your saved hardware

Desktop app links require the app to be installed. If nothing opens, LocalClaw will show app-download and model-file fallbacks.

Catalogue quality
9.4/10
Catalogue speed
6.4/10
Model size
42.4 GB
Voices
Direct TTS, optional speaker reference conditioning and TTS continuation

Can A.X K2 Raon-Speech 21B-A3B run locally?

A.X K2 Raon-Speech 21B-A3B can run locally for offline speech-to-text. Use the verified setup options on this page; no terminal command is required to choose the right path.

CC-BY-NC-4.0 license. Review upstream restrictions before commercial use.

Upstream source
cloningmultilingualdialoguecontrollable

Audio profile

Cat. quality
9.4
Cat. speed
6.4
Audio
8.4

Best fit

A.X K2 Raon-Speech 21B-A3B is best for offline transcription, speech indexing and local voice pipelines.

Hardware: gpu

Model details

Type
Local ASR model
Family
raon
Latency
medium
Formats
transformerssafetensors
Languages
en, ko
Context
21.2B total / 3.5B active MoE, ~42.4GB BF16 weights; 80GB GPU or suitable multi-GPU setup recommended

Install locally

01
Check runtimeConfirm the backend supports transformers, safetensors on your machine.
02
Open recommended setupUse the app and model links above. LocalClaw does not expose a terminal command.
03
Test locallyRun a short private audio prompt before moving into production workflows.

Good for

  • speech-to-text transcription
  • GPU recommended local workflows
  • cloning, multilingual, dialogue

Watch before shipping

  • Validate pronunciation, latency and artifacts with your own voice samples.
  • Review the upstream license and acceptable-use notes.
  • Benchmark on your target CPU, Apple Silicon or GPU setup.

Related TTS and speech models

KRAFTON AI Raon-OpenTTS-1B Local TTS model · Q 9.3 · Speed 7.7 FireRed Team FireRedAudio Local ASR model · Q 9.5 · Speed 6.6 Boson AI Higgs TTS 3 4B Local TTS model · Q 9.7 · Speed 8 OpenMOSS / MOSI.AI MOSS-Transcribe-Diarize 0.9B Local ASR model · Q 9.4 · Speed 8.4 IBM Granite Team Granite Speech 4.1 2B Plus Local ASR model · Q 9.3 · Speed 7.8 OKESTRO AI Lab / Lee Junseok FastSLM-ASR Local ASR model · Q 9.2 · Speed 7.8 Microsoft Research VibeVoice ASR Local ASR model · Q 9.3 · Speed 7.5 OpenMOSS / MOSI.AI MOSS-Audio 4B Instruct Local ASR model · Q 9.2 · Speed 7.7
CompareBrowse all TTS models Local AIBrowse LLM models macOS appGet LocalClaw