README.md

metadata

license: apache-2.0
base_model: facebook/data2vec-audio-base-960h
tags:
  - generated_from_trainer
datasets:
  - librispeech_asr
metrics:
  - wer
model-index:
  - name: jjyaoao/Echotune_clean_test
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: librispeech_asr
          type: librispeech_asr
          config: clean
          split: test
          args: clean
        metrics:
          - name: Wer
            type: wer
            value: 0.037368222891566265

jjyaoao/Echotune_clean_test

This model is a fine-tuned version of facebook/data2vec-audio-base-960h on the librispeech_asr dataset. It achieves the following results on the evaluation set:

• Loss: 0.0679
• Wer Ortho: 0.0369
• Wer: 0.0374

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: 6e-05
• train_batch_size: 12
• eval_batch_size: 12
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 34246.8
• num_epochs: 12

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.0602	0.21	500	0.0476	0.0435	0.0439
0.0478	0.42	1000	0.0436	0.0411	0.0414
0.0492	0.63	1500	0.0443	0.0412	0.0415
0.0426	0.84	2000	0.0439	0.0401	0.0403
0.0386	1.05	2500	0.0445	0.0391	0.0395
0.0409	1.26	3000	0.0438	0.0394	0.0399
0.0437	1.47	3500	0.0444	0.0389	0.0393
0.0349	1.68	4000	0.0450	0.0392	0.0396
0.0469	1.89	4500	0.0442	0.0374	0.0378
0.033	2.1	5000	0.0454	0.0359	0.0363
0.0395	2.31	5500	0.0462	0.0363	0.0367
0.0321	2.52	6000	0.0457	0.0365	0.0369
0.0385	2.73	6500	0.0455	0.0355	0.0358
0.0378	2.94	7000	0.0449	0.0361	0.0366
0.0435	3.15	7500	0.0440	0.0355	0.0360
0.0436	3.36	8000	0.0466	0.0339	0.0344
0.0394	3.57	8500	0.0480	0.0345	0.0350
0.0448	3.78	9000	0.0478	0.0338	0.0342
0.0451	3.99	9500	0.0460	0.0355	0.0361
0.035	4.2	10000	0.0485	0.0369	0.0374
0.0387	4.41	10500	0.0487	0.0358	0.0362
0.0479	4.62	11000	0.0496	0.0363	0.0368
0.0456	4.83	11500	0.0491	0.0359	0.0365
0.0372	5.04	12000	0.0507	0.0355	0.0360
0.0395	5.25	12500	0.0526	0.0353	0.0356
0.0323	5.46	13000	0.0515	0.0368	0.0373
0.0354	5.67	13500	0.0524	0.0338	0.0343
0.031	5.88	14000	0.0531	0.0349	0.0357
0.0295	6.09	14500	0.0560	0.0344	0.0349
0.032	6.31	15000	0.0564	0.0364	0.0369
0.0462	6.52	15500	0.0548	0.0358	0.0365
0.0467	6.73	16000	0.0562	0.0347	0.0352
0.0437	6.94	16500	0.0573	0.0354	0.0359
0.0357	7.15	17000	0.0561	0.0359	0.0362
0.0297	7.36	17500	0.0602	0.0347	0.0351
0.0388	7.57	18000	0.0552	0.0341	0.0345
0.0392	7.78	18500	0.0533	0.0326	0.0331
0.0419	7.99	19000	0.0535	0.0343	0.0349
0.0326	8.2	19500	0.0614	0.0374	0.0378
0.0423	8.41	20000	0.0585	0.0341	0.0346
0.0326	8.62	20500	0.0586	0.0356	0.0362
0.0448	8.83	21000	0.0637	0.0371	0.0375
0.0763	9.04	21500	0.0607	0.0359	0.0364
0.0317	9.25	22000	0.0635	0.0400	0.0405
0.0326	9.46	22500	0.0603	0.0368	0.0372
0.0393	9.67	23000	0.0665	0.0380	0.0385
0.0341	9.88	23500	0.0664	0.0408	0.0413
0.0351	10.09	24000	0.0638	0.0384	0.0388
0.0412	10.3	24500	0.0687	0.0380	0.0384
0.0359	10.51	25000	0.0634	0.0379	0.0385
0.047	10.72	25500	0.0652	0.0373	0.0378
0.0346	10.93	26000	0.0671	0.0390	0.0396
0.0366	11.14	26500	0.0664	0.0387	0.0393
0.0359	11.35	27000	0.0669	0.0369	0.0374
0.0366	11.56	27500	0.0705	0.0358	0.0364
0.054	11.77	28000	0.0659	0.0383	0.0390
0.0335	11.98	28500	0.0679	0.0369	0.0374

Framework versions

• Transformers 4.32.0.dev0
• Pytorch 2.0.0+cu117
• Datasets 2.13.1
• Tokenizers 0.13.3