Multi-talker Automatic Speech Recognition on Libri3Mix Clean (dev)
8.6WERID-30.D
Evaluation Results
| Method | Links | |
|---|---|---|
| ID-30.DLLM usage=With LLMs2026.03 | 8.6 | |
| ID-20.DLLM usage=With LLMs2026.03 | 9.8 | |
| w/o decoder: CTCLLM Usage=Without LLMs, Encoder=SSL2026.03 | 12.7 | |
| GEncSep w/o decoderLLM usage=Without LLMs, SSL for speech encoder=true, Alignment protocol=serialized CTC2026.03 | 12.7 | |
| GEncSepLLM Usage=Without LLMs, Encoder=SSL2026.03 | 13.3 | |
| GEncSepLLM usage=Without LLMs, SSL for speech encoder=true2026.03 | 13.3 | |
| ID-8LLM Usage=With LLMs2026.03 | 13.7 | |
| Training from ScratchLLM Usage=Without LLMs, Encoder=SSL2026.03 | 15 | |
| Training from ScratchLLM usage=Without LLMs, SSL for speech encoder=true2026.03 | 15 | |
| SOP-Llama-3BLLM Usage=With LLMs2026.03 | 17 | |
| SOP-Llama-3BLLM usage=With LLMs, Backbone=Llama-3B, Protocol=SOP2026.03 | 17 | |
| SOP-Llama-1BLLM Usage=With LLMs2026.03 | 20.8 | |
| SOP-Llama-1BLLM usage=With LLMs, Backbone=Llama-1B, Protocol=SOP2026.03 | 20.8 | |
| SOT-Llama-1BLLM Usage=With LLMs2026.03 | 21.5 | |
| SOT-Llama-1BLLM usage=With LLMs, Backbone=Llama-1B, Protocol=SOT2026.03 | 21.5 | |
| SOT-Llama-3BLLM Usage=With LLMs2026.03 | 22.3 | |
| SOT-Llama-3BLLM usage=With LLMs, Backbone=Llama-3B, Protocol=SOT2026.03 | 22.3 |