axlstm-cat

71.4% passed tests (439 passed / 176 failed).

Tests overview

Topic

Passed Tests

Correctness classification

40.0%

Correctness distribution

67.5%

Correctness speaker average

58.3%

Correctness speaker ranking

25.0%

Fairness accent

100.0%

Fairness language

100.0%

Fairness linguistic sentiment

100.0%

Fairness pitch

96.3%

Fairness sex

97.2%

Robustness background noise

35.0%

Robustness low quality phone

80.0%

Robustness recording condition

0.0%

Robustness simulated recording condition

0.0%

Robustness small changes

32.0%

Robustness spectral tilt

65.0%

Model parameters

Entry

Value

model

{‘architecture’: ‘axlstm’, ‘embeddings’: ‘axlstm-official/axlstm_base_200_16x4_8x128_fp16_r1’}

data

[‘msppodcast’]

sampling_rate

16000

quantization

False

Model information

Entry

Value

Author

aderington

Date

2024-10-17

Name

onnx

Subgroup

ser.basic-4.axlstm

Version

1.0.0

Data

{‘train’: {‘msppodcast’: {‘version’: ‘2.6.1’, ‘tables’: [‘emotion.categories.train.gold_standard’]}}, ‘dev’: {‘msppodcast’: {‘version’: ‘2.6.1’, ‘tables’: [‘emotion.categories.dev.gold_standard’]}}, ‘config’: {‘datasets’: {‘msppodcast-train-dev’: {‘name’: ‘msppodcast’, ‘version’: ‘2.6.1’, ‘tables’: {‘train’: ‘emotion.categories.train.gold_standard’, ‘dev’: ‘emotion.categories.dev.gold_standard’}}, ‘msppodcast-test-1’: {‘name’: ‘msppodcast’, ‘version’: ‘2.6.1’, ‘tables’: {‘test’: ‘emotion.categories.test-1.gold_standard’}}, ‘msppodcast-test-2’: {‘name’: ‘msppodcast’, ‘version’: ‘2.6.1’, ‘tables’: {‘test’: ‘emotion.categories.test-2.gold_standard’}}}}}

Model

{‘basemodel’: ‘axlstm-official/axlstm_base_200_16x4_8x128_fp16_r1’, ‘img_size’: (200, 80), ‘patch_size’: (4, 16), ‘frequency_first’: False, ‘expansion_factor’: 2, ‘alternation’: ‘bidirectional’, ‘depth_multiplier’: 1, ‘depth’: 12, ‘num_heads’: 12, ‘embed_dim’: 768, ‘freq_dim’: 5, ‘gradient_accum_steps’: 1, ‘use_fp16’: True, ‘per_device_train_batch_size’: 1024, ‘per_device_eval_batch_size’: 1024, ‘learning_rate’: 0.0001, ‘n_epochs’: 500, ‘min_length_train’: 0.5, ‘metric_for_eval’: ‘combined’, ‘main_metrics’: {‘emotion’: ‘UAR’}, ‘sampling_rate’: 16000, ‘seed’: 1, ‘tasks’: [‘emotion’], ‘labels’: {‘emotion’: [‘anger’, ‘happiness’, ‘neutral’, ‘sadness’]}, ‘task2problemtype’: {‘emotion’: ‘classification’}, ‘task2postprocessing’: {‘emotion’: None}, ‘task2lossweight’: {‘emotion’: 1}}