Table 2 Model performance comparison.

From: Adaptive context biasing in transformer-based ASR systems

Model

Dev

Test

CER%

CERR%

CER%

CERR%

Baseline

5.92

\(-6.47\)

6.55

\(-8.89\)

whisper-tiny28

29.06

\(-422.66\)

31.98

\(-432.11\)

whisper-large-v328

7.49

\(-34.71\)

8.08

\(-34.44\)

Wav2vec229

5.52

0.0072

5.87

2.32

CopyNE30

5.59

\(-0.0053\)

6.35

\(-0.056\)

Detector (ours)

5.56

–

6.01

–

Back to article page