Speech Recognition on In-House dataset
0.0305CERG1 + KOO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| G1 + KOODecoding type=Max{*}2026.01 | 0.0305 | 1.69 | |
| B1 + IOO-BDecoding type={0, 1}2026.01 | 0.0306 | 1.65 | |
| B1 + IOO-BDecoding type={0, 1, 2}2026.01 | 0.0306 | 1.54 | |
| G1 + KOODecoding type=Min{*}2026.01 | 0.0306 | 1.67 | |
| B1 + SWDDecoding type={0, ± 1}2026.01 | 0.0308 | 1.47 | |
| A1 + 5-gramDecoding type=Dense CTC FST2026.01 | 0.0309 | 1 | |
| B1 + Speech-LLaMADecoding type=Averaging2026.01 | 0.031 | 1.29 | |
| B1 + LSDDecoding type=0.99 blank threshold2026.01 | 0.0312 | 1.18 | |
| B2 + IOO-BDecoding type={0, 1}2026.01 | 0.0313 | 0.43 | |
| AED ZipformerDecoding type=AED Greedy Search2026.01 | 0.0321 | 0.87 | |
| CTC ZipformerDecoding type=CTC Greedy Search2026.01 | 0.0327 | 3.45 | |
| A2 + 5-gramDecoding type=Dense AED FST2026.01 | 0.0356 | 0.51 | |
| G1 + IOO-NBDecoding type=Max{*}2026.01 | 0.0375 | 1.75 | |
| G1 + IOO-NBDecoding type={*}2026.01 | 0.0378 | 1.72 | |
| B1 + PolyVoiceDecoding type=Discarding2026.01 | 0.0471 | 1.99 | |
| Proposed (streaming LLM-based ASR)Model type=decoder-only, Streaming=false2026.01 | 0.067 | — | |
| Proposed (streaming LLM-based ASR)Model type=decoder-only, Streaming=true2026.01 | 0.076 | — | |
| Baseline-non-streamModel type=encoder-decoder, Streaming=false2026.01 | 0.08 | — | |
| Baseline-streamModel type=encoder-decoder, Streaming=true2026.01 | 0.096 | — |