wavlm

88.3% passed tests (363 passed / 48 failed).

Tests overview

Topic

Passed Tests

Correctness consistency

59.5%

Correctness distribution

33.3%

Correctness regression

66.7%

Correctness speaker average

66.7%

Correctness speaker ranking

100.0%

Fairness accent

97.6%

Fairness language

80.0%

Fairness linguistic sentiment

99.0%

Fairness pitch

93.3%

Fairness sex

100.0%

Robustness background noise

62.5%

Robustness low quality phone

100.0%

Robustness recording condition

50.0%

Robustness simulated recording condition

33.3%

Robustness small changes

100.0%

Robustness spectral tilt

75.0%

Model parameters

Entry

Value

data

MSP-Podcast_Challenge

model

MSP-Podcast Dim Baseline

sampling_rate

16000

attn_pool

True

quantization

False

Model information

Entry

Value

Author

aderington

Date

2024-03-05

Name

onnx

Subgroup

ser.adv.wavlm

Version

1.0.1