hubert-large-hre-v1

This model is a fine-tuned version of facebook/hubert-large-ls960-ft on the hre-audio-dataset8 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7225
• Cer Ortho: 53.9483
• Cer: 46.5317

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer Ortho	Cer
3.3539	0.4608	100	3.2333	99.0488	99.0248
2.1675	0.9217	200	2.0411	84.2247	82.1343
1.2398	1.3825	300	1.2868	69.8313	66.8261
1.0372	1.8433	400	1.0814	62.0603	60.5520
0.9143	2.3041	500	0.9490	64.1780	59.4296
0.8814	2.7650	600	0.8401	61.5757	56.4857
0.7726	3.2258	700	0.8606	58.6504	55.9522
0.8116	3.6866	800	0.7670	67.9648	54.9402
0.6826	4.1475	900	0.7524	67.7315	54.1490
0.5953	4.6083	1000	0.7812	66.7085	53.5051
0.6733	5.0691	1100	0.8268	63.0294	54.0202
0.6596	5.5300	1200	0.7109	62.9935	52.6403
0.5065	5.9908	1300	0.7529	65.4702	51.6651
0.4880	6.4516	1400	0.7643	65.0395	51.5731
0.5539	6.9124	1500	0.7285	65.6676	51.8123
0.5446	7.3733	1600	0.7215	62.7782	51.1316
0.4763	7.8341	1700	0.7669	64.2498	51.2971
0.4665	8.2949	1800	0.6920	63.8729	50.1564
0.4189	8.7558	1900	0.6964	64.4472	49.9172
0.4124	9.2166	2000	0.7164	58.8119	49.9540
0.3157	9.6774	2100	0.6939	56.5506	49.9356
0.3570	10.1382	2200	0.7459	54.9174	49.4756
0.3523	10.5991	2300	0.6848	50.9512	48.7029
0.3624	11.0599	2400	0.6860	50.6281	48.6293
0.3054	11.5207	2500	0.7821	51.3640	48.8684
0.3190	11.9816	2600	0.8238	52.4587	49.2732
0.3675	12.4424	2700	0.7947	57.9864	49.3284
0.4182	12.9032	2800	0.7326	57.7889	48.3533
0.2994	13.3641	2900	0.7606	55.7251	48.4085
0.2666	13.8249	3000	0.7204	56.0660	47.6909
0.2769	14.2857	3100	0.7308	54.8636	47.3229
0.2843	14.7465	3200	0.7531	56.6403	47.8381
0.2587	15.2074	3300	0.7166	56.1378	47.3965
0.2386	15.6682	3400	0.6969	52.9433	47.2125
0.2284	16.1290	3500	0.7046	51.2563	46.8813
0.1806	16.5899	3600	0.7277	52.7997	47.1205
0.2266	17.0507	3700	0.6840	53.8227	46.5133
0.1728	17.5115	3800	0.7269	53.8586	46.5317
0.1993	17.9724	3900	0.7186	54.5585	46.5869
0.1944	18.4332	4000	0.7228	54.5226	46.3109
0.1923	18.8940	4100	0.7402	54.7739	46.8077
0.2191	19.3548	4200	0.7298	53.9842	46.5501
0.1717	19.8157	4300	0.7230	54.0022	46.5133
0.1761	20.0	4340	0.7225	53.9483	46.5317

Framework versions

• Transformers 5.13.1
• Pytorch 2.11.0+cu128
• Datasets 2.18.0
• Tokenizers 0.22.2