wav2vec2-large-xlsr-53-common_voice-ja-demo-kana-only

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the MOZILLA-FOUNDATION/COMMON_VOICE_13_0 - JA dataset. It achieves the following results on the evaluation set:

• Loss: 0.6985
• Wer: 0.9998
• Cer: 0.3126

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: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	0.2660	100	6.8309	1.0	0.9999
No log	0.5319	200	4.1299	1.0	0.9999
No log	0.7979	300	3.9930	1.0	0.9869
No log	1.0638	400	2.0400	1.0	0.5876
7.1041	1.3298	500	1.0610	1.0	0.4309
7.1041	1.5957	600	0.8837	1.0	0.3955
7.1041	1.8617	700	0.7706	0.9998	0.3791
7.1041	2.1277	800	0.7662	1.0	0.3816
7.1041	2.3936	900	0.7621	1.0	0.3790
0.803	2.6596	1000	0.6969	1.0	0.3626
0.803	2.9255	1100	0.6736	1.0	0.3573
0.803	3.1915	1200	0.6823	0.9998	0.3544
0.803	3.4574	1300	0.6360	1.0	0.3460
0.803	3.7234	1400	0.6504	1.0	0.3443
0.5675	3.9894	1500	0.6247	1.0	0.3414
0.5675	4.2553	1600	0.6397	0.9998	0.3425
0.5675	4.5213	1700	0.6589	1.0	0.3439
0.5675	4.7872	1800	0.6345	1.0	0.3449
0.5675	5.0532	1900	0.6522	0.9996	0.3380
0.4421	5.3191	2000	0.6293	1.0	0.3372
0.4421	5.5851	2100	0.6096	1.0	0.3342
0.4421	5.8511	2200	0.6108	1.0	0.3321
0.4421	6.1170	2300	0.6200	1.0	0.3354
0.4421	6.3830	2400	0.6413	1.0	0.3341
0.3699	6.6489	2500	0.6303	0.9996	0.3359
0.3699	6.9149	2600	0.6013	1.0	0.3308
0.3699	7.1809	2700	0.6343	1.0	0.3286
0.3699	7.4468	2800	0.6208	0.9998	0.3260
0.3699	7.7128	2900	0.6095	0.9998	0.3287
0.3146	7.9787	3000	0.6058	0.9996	0.3266
0.3146	8.2447	3100	0.6613	0.9996	0.3251
0.3146	8.5106	3200	0.6539	1.0	0.3244
0.3146	8.7766	3300	0.6331	1.0	0.3264
0.3146	9.0426	3400	0.6436	1.0	0.3228
0.2576	9.3085	3500	0.6329	1.0	0.3235
0.2576	9.5745	3600	0.6315	0.9998	0.3197
0.2576	9.8404	3700	0.6281	0.9998	0.3203
0.2576	10.1064	3800	0.6696	0.9996	0.3196
0.2576	10.3723	3900	0.6630	0.9996	0.3199
0.2201	10.6383	4000	0.6781	1.0	0.3203
0.2201	10.9043	4100	0.6531	1.0	0.3196
0.2201	11.1702	4200	0.6763	0.9998	0.3193
0.2201	11.4362	4300	0.6785	1.0	0.3184
0.2201	11.7021	4400	0.6664	0.9998	0.3179
0.1931	11.9681	4500	0.6682	0.9998	0.3184
0.1931	12.2340	4600	0.6800	0.9998	0.3168
0.1931	12.5	4700	0.6925	1.0	0.3162
0.1931	12.7660	4800	0.7047	1.0	0.3145
0.1931	13.0319	4900	0.6919	0.9998	0.3147
0.1694	13.2979	5000	0.6999	0.9998	0.3142
0.1694	13.5638	5100	0.6995	1.0	0.3134
0.1694	13.8298	5200	0.6917	0.9998	0.3134
0.1694	14.0957	5300	0.6963	0.9998	0.3129
0.1694	14.3617	5400	0.6961	0.9998	0.3128
0.1548	14.6277	5500	0.6964	1.0	0.3129
0.1548	14.8936	5600	0.6984	0.9998	0.3127

Framework versions

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