List your company

Speech Recognition on LibriSpeech (test-clean)

0.59BWERPhi-4-Multimodal + RLBR
-2.84420.335543.51566.6945Jan 19, 2026

Evaluation results

MethodLinks
Phi-4-Multimodal + RLBRN (Biasing list size)=1002026.010.590.820.85
Phi-4-Multimodal + SFTN (Biasing list size)=1002026.011.060.950.93
Phi-4-Multimodal + RLBRN (Biasing list size)=5002026.011.090.960.95
Phi-4-Multimodal + RLBRN (Biasing list size)=10002026.011.361.071.03
Phi-4-Multimodal + SFTN (Biasing list size)=5002026.011.71.060.98
Phi-4-Multimodal + SFTN (Biasing list size)=10002026.012.381.140.99
Dynamic VocabularyN (Biasing list size)=1002026.012.81.81.7
Dynamic VocabularyN (Biasing list size)=5002026.013.11.921.8
Dynamic VocabularyN (Biasing list size)=10002026.013.32.011.9
CTC-Assisted LLMN (Biasing list size)=1002026.013.671.271
CTC-Assisted LLMN (Biasing list size)=5002026.013.921.331.03
CTC-Assisted LLMN (Biasing list size)=10002026.014.161.331
Phi-4-MultimodalN (Biasing list size)=1002026.015.891.480.92
Phi-4-MultimodalN (Biasing list size)=02026.017.441.670.94
Phi-4-Multimodal + SFTN (Biasing list size)=02026.017.491.720.99
Phi-4-Multimodal + RLBRN (Biasing list size)=02026.017.821.720.95
CTC-Assisted LLMN (Biasing list size)=02026.019.331.961.11
Dynamic VocabularyN (Biasing list size)=02026.0113.83.161.9
Phi-4-MultimodalN (Biasing list size)=5002026.0137.87116126
Phi-4-MultimodalN (Biasing list size)=10002026.0186.44201215