README.md

metadata

license: apache-2.0
base_model: facebook/wav2vec2-large-xlsr-53
tags:
  - generated_from_trainer
datasets:
  - common_voice
metrics:
  - wer
model-index:
  - name: wav2vec2-large-xlsr53-zh-cn-subset20-colab
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: common_voice
          type: common_voice
          config: zh-CN
          split: test[:20%]
          args: zh-CN
        metrics:
          - name: Wer
            type: wer
            value: 0.9503424657534246

wav2vec2-large-xlsr53-zh-cn-subset20-colab

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

• Loss: 2.0566
• Wer: 0.9503
• Cer: 0.3333

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: 13
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 26
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	1.9	400	6.7551	1.0	1.0
34.7845	3.81	800	6.4563	1.0	1.0
6.4358	5.71	1200	4.2319	1.0074	0.7454
4.2052	7.62	1600	2.6538	1.0200	0.5562
2.3906	9.52	2000	2.3565	1.0063	0.5147
2.3906	11.43	2400	2.1287	0.9863	0.4822
1.93	13.33	2800	1.9585	0.9812	0.4528
1.6322	15.24	3200	1.8771	0.9937	0.4381
1.3629	17.14	3600	1.8405	0.9926	0.4242
1.166	19.05	4000	1.7674	0.9989	0.4140
1.166	20.95	4400	1.7879	0.9795	0.4047
0.9915	22.86	4800	1.7597	1.0126	0.4080
0.8517	24.76	5200	1.7726	0.9829	0.3966
0.7143	26.67	5600	1.7623	0.9732	0.3863
0.6267	28.57	6000	1.8164	0.9720	0.3863
0.6267	30.48	6400	1.8136	0.9680	0.3801
0.5389	32.38	6800	1.8696	0.9652	0.3812
0.4764	34.29	7200	1.8625	0.9663	0.3744
0.4095	36.19	7600	1.8868	0.9618	0.3683
0.3594	38.1	8000	1.8834	0.9623	0.3699
0.3594	40.0	8400	1.9155	0.9589	0.3670
0.3064	41.9	8800	1.9268	0.9652	0.3688
0.2825	43.81	9200	1.9527	0.9697	0.3674
0.2524	45.71	9600	1.9726	0.9686	0.3617
0.2272	47.62	10000	1.9594	0.9629	0.3619
0.2272	49.52	10400	1.9799	0.9635	0.3607
0.2042	51.43	10800	2.0175	0.9669	0.3582
0.1975	53.33	11200	2.0246	0.9589	0.3571
0.1827	55.24	11600	2.0535	0.9703	0.3600
0.1677	57.14	12000	2.0458	0.9583	0.3555
0.1677	59.05	12400	2.0893	0.9572	0.3583
0.1626	60.95	12800	2.0729	0.9600	0.3557
0.155	62.86	13200	2.0706	0.9572	0.3538
0.1456	64.76	13600	2.0761	0.9532	0.3553
0.1337	66.67	14000	2.0349	0.9589	0.3474
0.1337	68.57	14400	2.0844	0.9549	0.3484
0.1274	70.48	14800	2.0874	0.9578	0.3505
0.1198	72.38	15200	2.0813	0.9526	0.3473
0.1164	74.29	15600	2.0866	0.9498	0.3473
0.1105	76.19	16000	2.0688	0.9486	0.3421
0.1105	78.1	16400	2.0854	0.9498	0.3431
0.1053	80.0	16800	2.0749	0.9503	0.3414
0.1	81.9	17200	2.0622	0.9543	0.3407
0.0977	83.81	17600	2.0678	0.9532	0.3396
0.0906	85.71	18000	2.0650	0.9515	0.3383
0.0906	87.62	18400	2.0631	0.9492	0.3378
0.0867	89.52	18800	2.0633	0.9521	0.3365
0.0836	91.43	19200	2.0606	0.9532	0.3346
0.0819	93.33	19600	2.0671	0.9538	0.3355
0.0768	95.24	20000	2.0661	0.9509	0.3338
0.0768	97.14	20400	2.0564	0.9498	0.3335
0.0752	99.05	20800	2.0566	0.9503	0.3333

Framework versions

• Transformers 4.32.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.14.3
• Tokenizers 0.13.3