Edit model card

wav2vec2-large-xls-r-300m-hi-cv8

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - HI dataset. It achieves the following results on the evaluation set:

  • Loss: 0.6510
  • Wer: 0.3179

Evaluation Commands

  1. To evaluate on mozilla-foundation/common_voice_8_0 with test split

python eval.py --model_id DrishtiSharma/wav2vec2-large-xls-r-300m-hi-cv8 --dataset mozilla-foundation/common_voice_8_0 --config hi --split test --log_outputs

  1. To evaluate on speech-recognition-community-v2/dev_data

python eval.py --model_id DrishtiSharma/wav2vec2-large-xls-r-300m-hi-cv8 --dataset speech-recognition-community-v2/dev_data --config hi --split validation --chunk_length_s 10 --stride_length_s 1

Note: Hindi language not found in speech-recognition-community-v2/dev_data

Training hyperparameters

The following hyperparameters were used during training:

  • train_batch_size: 16
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 2000
  • num_epochs: 50
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
12.5576 1.04 200 6.6594 1.0
4.4069 2.07 400 3.6011 1.0
3.4273 3.11 600 3.3370 1.0
2.1108 4.15 800 1.0641 0.6562
0.8817 5.18 1000 0.7178 0.5172
0.6508 6.22 1200 0.6612 0.4839
0.5524 7.25 1400 0.6458 0.4889
0.4992 8.29 1600 0.5791 0.4382
0.4669 9.33 1800 0.6039 0.4352
0.4441 10.36 2000 0.6276 0.4297
0.4172 11.4 2200 0.6183 0.4474
0.3872 12.44 2400 0.5886 0.4231
0.3692 13.47 2600 0.6448 0.4399
0.3385 14.51 2800 0.6344 0.4075
0.3246 15.54 3000 0.5896 0.4087
0.3026 16.58 3200 0.6158 0.4016
0.284 17.62 3400 0.6038 0.3906
0.2682 18.65 3600 0.6165 0.3900
0.2577 19.69 3800 0.5754 0.3805
0.2509 20.73 4000 0.6028 0.3925
0.2426 21.76 4200 0.6335 0.4138
0.2346 22.8 4400 0.6128 0.3870
0.2205 23.83 4600 0.6223 0.3831
0.2104 24.87 4800 0.6122 0.3781
0.1992 25.91 5000 0.6467 0.3792
0.1916 26.94 5200 0.6277 0.3636
0.1835 27.98 5400 0.6317 0.3773
0.1776 29.02 5600 0.6124 0.3614
0.1751 30.05 5800 0.6475 0.3628
0.1662 31.09 6000 0.6266 0.3504
0.1584 32.12 6200 0.6347 0.3532
0.1494 33.16 6400 0.6636 0.3491
0.1457 34.2 6600 0.6334 0.3507
0.1427 35.23 6800 0.6397 0.3442
0.1397 36.27 7000 0.6468 0.3496
0.1283 37.31 7200 0.6291 0.3416
0.1255 38.34 7400 0.6652 0.3461
0.1195 39.38 7600 0.6587 0.3342
0.1169 40.41 7800 0.6478 0.3319
0.1126 41.45 8000 0.6280 0.3291
0.1112 42.49 8200 0.6434 0.3290
0.1069 43.52 8400 0.6542 0.3268
0.1027 44.56 8600 0.6536 0.3239
0.0993 45.6 8800 0.6622 0.3257
0.0973 46.63 9000 0.6572 0.3192
0.0911 47.67 9200 0.6522 0.3175
0.0897 48.7 9400 0.6521 0.3200
0.0905 49.74 9600 0.6510 0.3179

Framework versions

  • Transformers 4.16.2
  • Pytorch 1.10.0+cu111
  • Datasets 1.18.3
  • Tokenizers 0.11.0
Downloads last month
11
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train DrishtiSharma/wav2vec2-large-xls-r-300m-hi-cv8

Evaluation results