SimulS2ST-Omni model assets
Model assets for SimulS2ST-Omni, accepted by EMNLP 2026.
Layout
| Directory | Contents | Approximate size |
|---|---|---|
offline/ |
Fully merged OmniTalker model, offline talker components, processor, and tokenizer files | 13 GB |
simuls2st_adapter/ |
Simultaneous S2ST thinker/talker LoRA adapters, talker components, and tokenizer files | 893 MB |
voicebox/ |
VoiceBox, Vocos, and voicebox_config.json |
2.3 GB |
dualcodec/ |
DualCodec weights and W2V-BERT normalization statistics | 321 MB |
w2v/ |
facebook/w2v-bert-2.0 feature model used by the codec tokenizer |
2.2 GB |
The root simuls2st_config.json records the canonical relative path for every
runtime component. MANIFEST.sha256 verifies all runtime assets.
Download
hf download HA-SA-ki/SimulS2ST-Omni \
--local-dir models/SimulS2ST-Omni
Runtime paths
models/SimulS2ST-Omni/offline
models/SimulS2ST-Omni/simuls2st_adapter
models/SimulS2ST-Omni/voicebox/voicebox.safetensors
models/SimulS2ST-Omni/voicebox/vocos.safetensors
models/SimulS2ST-Omni/voicebox/voicebox_config.json
models/SimulS2ST-Omni/dualcodec/dualcodec.safetensors
models/SimulS2ST-Omni/dualcodec/w2v_bert_stats.pt
models/SimulS2ST-Omni/w2v
For offline inference, use offline/ as both the model and checkpoint path.
For simultaneous streaming inference, use offline/ as the base model and
simuls2st_adapter/ as the adapter/checkpoint path.
License
The SimulS2ST-Omni model weights are released under the Apache License 2.0.
Upload
Upload large files with the Hugging Face CLI rather than a regular Git push:
hf upload-large-folder HA-SA-ki/SimulS2ST-Omni . --repo-type model