wav2vec-jdrt-RTSplit_and_SpeakerSplitModel-0107-5

This model is a fine-tuned version of jonatasgrosman/wav2vec2-large-xlsr-53-japanese on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0032
• Wer: 0.1835
• Cer: 0.1557

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: 5e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 35

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
14.5102	1.0	60	13.5418	1.0011	1.2966
6.2989	2.0	120	4.6126	1.0	0.8736
2.208	3.0	180	1.7112	0.9993	0.8160
1.098	4.0	240	0.8765	0.8410	0.5698
0.8201	5.0	300	0.6964	0.8221	0.5340
0.7499	6.0	360	0.6299	0.8217	0.5305
0.6753	7.0	420	0.5998	0.7691	0.4482
0.6003	8.0	480	0.5502	0.7394	0.4564
0.5732	9.0	540	0.5047	0.7098	0.3906
0.5404	10.0	600	0.4694	0.6679	0.3283
0.4889	11.0	660	0.3979	0.6379	0.3017
0.4401	12.0	720	0.3255	0.5849	0.2792
0.4295	13.0	780	0.2853	0.5044	0.2772
0.3216	14.0	840	0.2204	0.4511	0.2234
0.2583	15.0	900	0.1492	0.3929	0.2130
0.226	16.0	960	0.1007	0.2817	0.1690
0.2304	17.0	1020	0.0694	0.2439	0.1699
0.1487	18.0	1080	0.0471	0.2142	0.1758
0.1045	19.0	1140	0.0305	0.2168	0.1686
0.1104	20.0	1200	0.0256	0.2072	0.1625
0.094	21.0	1260	0.0226	0.2272	0.1760
0.0987	22.0	1320	0.0129	0.2013	0.1900
0.0753	23.0	1380	0.0110	0.2053	0.1786
0.0544	24.0	1440	0.0091	0.1909	0.1858
0.0684	25.0	1500	0.0083	0.1901	0.1728
0.0723	26.0	1560	0.0083	0.2027	0.1854
0.061	27.0	1620	0.0061	0.2020	0.1779
0.0635	28.0	1680	0.0059	0.1964	0.1818
0.0336	29.0	1740	0.0048	0.1887	0.1574
0.0455	30.0	1800	0.0036	0.1842	0.1694
0.0672	31.0	1860	0.0038	0.1838	0.1507
0.0315	32.0	1920	0.0033	0.1853	0.1555
0.0466	33.0	1980	0.0033	0.1827	0.1569
0.0491	34.0	2040	0.0035	0.1835	0.1556
0.0315	35.0	2100	0.0032	0.1835	0.1557

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.14.6
• Tokenizers 0.15.0