Hubert_noisy_common_voice_phonemes_debug

This model is a fine-tuned version of rinna/japanese-hubert-base on the ORIGINAL_NOISY_COMMON_VOICE - JA dataset. It achieves the following results on the evaluation set:

• Loss: 0.9125
• Wer: 1.0222
• Cer: 0.3103

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: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 12500
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	0.2660	100	12.2968	1.0717	1.0679
No log	0.5319	200	5.9423	1.0	0.9813
No log	0.7979	300	5.4201	1.0	0.9813
No log	1.0638	400	4.9840	1.0	0.9813
6.4953	1.3298	500	4.4992	1.0	0.9813
6.4953	1.5957	600	4.0212	1.0	0.9813
6.4953	1.8617	700	3.5973	1.0	0.9813
6.4953	2.1277	800	3.2996	1.0	0.9813
6.4953	2.3936	900	3.1686	1.0	0.9813
3.442	2.6596	1000	3.0725	1.0	0.9813
3.442	2.9255	1100	2.9187	1.0	0.9813
3.442	3.1915	1200	2.6014	1.0	0.8917
3.442	3.4574	1300	2.1700	1.0	0.6548
3.442	3.7234	1400	1.7176	1.0	0.4492
2.3862	3.9894	1500	1.5003	1.0	0.4197
2.3862	4.2553	1600	1.3507	1.0	0.4027
2.3862	4.5213	1700	1.2036	1.0	0.3701
2.3862	4.7872	1800	1.0972	1.0	0.3432
2.3862	5.0532	1900	0.9528	1.0	0.3108
1.2375	5.3191	2000	0.8881	1.0	0.2965
1.2375	5.5851	2100	0.8716	1.0	0.3024
1.2375	5.8511	2200	0.8150	1.0	0.2904
1.2375	6.1170	2300	0.7949	1.0	0.2875
1.2375	6.3830	2400	0.7734	1.0	0.2879
0.8538	6.6489	2500	0.7513	1.0	0.2839
0.8538	6.9149	2600	0.7448	1.0	0.2822
0.8538	7.1809	2700	0.7400	1.0	0.2804
0.8538	7.4468	2800	0.7283	1.0	0.2786
0.8538	7.7128	2900	0.7322	1.0	0.2809
0.7165	7.9787	3000	0.7111	1.0	0.2784
0.7165	8.2447	3100	0.7282	1.0	0.2858
0.7165	8.5106	3200	0.6960	1.0	0.2750
0.7165	8.7766	3300	0.7104	1.0	0.2811
0.7165	9.0426	3400	0.7289	1.0006	0.2790
0.6393	9.3085	3500	0.7068	1.0	0.2787
0.6393	9.5745	3600	0.7173	0.9999	0.2768
0.6393	9.8404	3700	0.6848	0.9963	0.2711
0.6393	10.1064	3800	0.7057	0.9954	0.2792
0.6393	10.3723	3900	0.7190	0.9975	0.2792
0.5993	10.6383	4000	0.7214	0.9946	0.2779
0.5993	10.9043	4100	0.7275	0.9931	0.2832
0.5993	11.1702	4200	0.6970	0.9902	0.2744
0.5993	11.4362	4300	0.7212	0.9946	0.2723
0.5993	11.7021	4400	0.7260	0.9915	0.2751
0.5646	11.9681	4500	0.7185	1.0111	0.2737
0.5646	12.2340	4600	0.7415	0.9968	0.2833
0.5646	12.5	4700	0.7404	0.9908	0.2779
0.5646	12.7660	4800	0.7145	0.9885	0.2727
0.5646	13.0319	4900	0.7319	1.0011	0.2719
0.5215	13.2979	5000	0.7503	0.9994	0.2726
0.5215	13.5638	5100	0.7200	1.0067	0.2710
0.5215	13.8298	5200	0.7043	0.9895	0.2746
0.5215	14.0957	5300	0.7587	1.0130	0.2760
0.5215	14.3617	5400	0.7453	0.9886	0.2792
0.4978	14.6277	5500	0.7269	1.0015	0.2754
0.4978	14.8936	5600	0.7381	0.9986	0.2728
0.4978	15.1596	5700	0.7658	1.0445	0.2747
0.4978	15.4255	5800	0.7593	1.0165	0.2758
0.4978	15.6915	5900	0.7959	1.0401	0.2799
0.4807	15.9574	6000	0.7533	1.0161	0.2784
0.4807	16.2234	6100	0.7566	0.9879	0.2775
0.4807	16.4894	6200	0.7418	0.9918	0.2784
0.4807	16.7553	6300	0.7968	0.9957	0.2811
0.4807	17.0213	6400	0.7728	1.0132	0.2754
0.4456	17.2872	6500	0.8130	1.0176	0.2794
0.4456	17.5532	6600	0.8082	1.0552	0.2850
0.4456	17.8191	6700	0.8325	1.0939	0.2797
0.4456	18.0851	6800	0.8033	0.9931	0.2804
0.4456	18.3511	6900	0.7595	1.0057	0.2801
0.4396	18.6170	7000	0.7648	1.0057	0.2816
0.4396	18.8830	7100	0.7651	0.9965	0.2818
0.4396	19.1489	7200	0.7942	1.0526	0.2821
0.4396	19.4149	7300	0.7584	1.0329	0.2865
0.4396	19.6809	7400	0.7743	1.0247	0.2839
0.4402	19.9468	7500	0.7724	0.9974	0.2782
0.4402	20.2128	7600	0.8211	1.0083	0.2819
0.4402	20.4787	7700	0.7944	0.9985	0.2845
0.4402	20.7447	7800	0.8000	1.0283	0.2809
0.4402	21.0106	7900	0.7961	1.0393	0.2848
0.4161	21.2766	8000	0.8153	1.0126	0.2868
0.4161	21.5426	8100	0.7890	1.0290	0.2848
0.4161	21.8085	8200	0.8137	0.9949	0.2876
0.4161	22.0745	8300	0.8160	1.0130	0.2883
0.4161	22.3404	8400	0.8261	0.9967	0.2843
0.4122	22.6064	8500	0.8360	1.0004	0.2872
0.4122	22.8723	8600	0.7974	0.9870	0.2845
0.4122	23.1383	8700	0.8509	1.0251	0.2959
0.4122	23.4043	8800	0.8392	1.0060	0.2996
0.4122	23.6702	8900	0.8572	1.0025	0.2960
0.4233	23.9362	9000	0.8738	1.0243	0.2959
0.4233	24.2021	9100	0.8740	1.0279	0.2897
0.4233	24.4681	9200	0.8348	1.0178	0.2910
0.4233	24.7340	9300	0.8519	1.0287	0.2965
0.4233	25.0	9400	0.8510	0.9975	0.3038
0.4072	25.2660	9500	0.8886	1.0440	0.2998
0.4072	25.5319	9600	0.9135	0.9960	0.3032
0.4072	25.7979	9700	0.8631	1.0018	0.3065
0.4072	26.0638	9800	0.8652	1.0216	0.2992
0.4072	26.3298	9900	0.8664	1.0366	0.2960
0.4149	26.5957	10000	0.8856	1.0248	0.3047
0.4149	26.8617	10100	0.8662	1.0223	0.2998
0.4149	27.1277	10200	0.9195	0.9953	0.3116
0.4149	27.3936	10300	0.9434	1.0148	0.3118
0.4149	27.6596	10400	0.8643	1.0126	0.3096
0.4264	27.9255	10500	0.9074	1.0062	0.3078
0.4264	28.1915	10600	0.8856	1.0497	0.3035
0.4264	28.4574	10700	0.8924	1.0676	0.3032
0.4264	28.7234	10800	0.9018	1.0203	0.3002
0.4264	28.9894	10900	0.9206	1.0573	0.3049
0.4091	29.2553	11000	0.8745	1.0294	0.3033
0.4091	29.5213	11100	0.8626	0.9920	0.3053
0.4091	29.7872	11200	0.9597	1.0218	0.3129

Framework versions

• Transformers 4.47.0.dev0
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3