Hugging Face's logo

Alvin-Nahabwe
/
w2v-bert-2.0-hausa-robust-1000hrs-v0.0

Automatic Speech Recognition
Transformers
Safetensors
Hausa
wav2vec2-bert
asr
speech
asr-africa
robust-fine-tuning
Generated from Trainer
Eval Results
Model card Files Community
w2v-bert-2.0-hausa-robust-1000hrs-v0.0 / README.md
Alvin-Nahabwe's picture
Alvin-Nahabwe
End of training
8382e7e verified
preview code
|
raw
history blame contribute delete
10.5 kB
metadata
library_name: transformers
language:
  - ha
license: mit
base_model: facebook/w2v-bert-2.0
tags:
  - asr
  - ha
  - wav2vec2-bert
  - speech
  - asr-africa
  - robust-fine-tuning
  - generated_from_trainer
datasets:
  - naijavoices/naijavoices-dataset
  - CLEAR-Global/Hausa-Synthetic-ASR-Dataset
  - google/fleurs
  - mozilla-foundation/common_voice_17_0
  - benjaminogbonna/nigerian_common_voice_dataset
metrics:
  - wer
model-index:
  - name: Wav2Vec2-BERT - Hausa - asr-africa
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: naijavoices/naijavoices-dataset
          type: naijavoices/naijavoices-dataset
        metrics:
          - name: Wer
            type: wer
            value: 0.11443126434701978
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: CLEAR-Global/Hausa-Synthetic-ASR-Dataset
          type: CLEAR-Global/Hausa-Synthetic-ASR-Dataset
        metrics:
          - name: Wer
            type: wer
            value: 0.11443126434701978
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: google/fleurs
          type: google/fleurs
        metrics:
          - name: Wer
            type: wer
            value: 0.11443126434701978
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: mozilla-foundation/common_voice_17_0
          type: mozilla-foundation/common_voice_17_0
        metrics:
          - name: Wer
            type: wer
            value: 0.11443126434701978
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: benjaminogbonna/nigerian_common_voice_dataset
          type: benjaminogbonna/nigerian_common_voice_dataset
        metrics:
          - name: Wer
            type: wer
            value: 0.11443126434701978

Wav2Vec2-BERT - Hausa - asr-africa

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the naijavoices/naijavoices-dataset, the CLEAR-Global/Hausa-Synthetic-ASR-Dataset, the google/fleurs, the mozilla-foundation/common_voice_17_0 and the benjaminogbonna/nigerian_common_voice_dataset datasets. It achieves the following results on the evaluation set:

  • Loss: 0.7709
  • Wer: 0.1144
  • Cer: 0.0356

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: 64
  • eval_batch_size: 32
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 256
  • optimizer: Use OptimizerNames.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_ratio: 0.01
  • num_epochs: 100.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Cer
0.2579 1.0 3863 0.3751 0.3166 0.0878
0.1668 2.0 7726 0.3403 0.2865 0.0801
0.1513 3.0 11589 0.3114 0.2763 0.0758
0.1411 4.0 15452 0.3138 0.2775 0.0773
0.1329 5.0 19315 0.2924 0.2577 0.0699
0.1295 6.0 23178 0.3022 0.2673 0.0733
0.1221 7.0 27041 0.2858 0.2571 0.0704
0.1208 8.0 30904 0.2983 0.2620 0.0718
0.1148 9.0 34767 0.2683 0.2491 0.0676
0.1128 10.0 38630 0.2647 0.2416 0.0648
0.1107 11.0 42493 0.2781 0.2548 0.0692
0.105 12.0 46356 0.2669 0.2399 0.0651
0.1036 13.0 50219 0.2748 0.2498 0.0689
0.1003 14.0 54082 0.2519 0.2328 0.0621
0.097 15.0 57945 0.2537 0.2304 0.0612
0.0949 16.0 61808 0.2671 0.2432 0.0652
0.0917 17.0 65671 0.2470 0.2275 0.0608
0.0894 18.0 69534 0.2664 0.2410 0.0652
0.0872 19.0 73397 0.2532 0.2289 0.0621
0.0845 20.0 77260 0.2590 0.2309 0.0628
0.0813 21.0 81123 0.2533 0.2235 0.0603
0.0791 22.0 84986 0.2486 0.2196 0.0590
0.0766 23.0 88849 0.2450 0.2153 0.0573
0.0737 24.0 92712 0.2391 0.2151 0.0577
0.0725 25.0 96575 0.2388 0.2103 0.0564
0.0669 26.0 100438 0.2347 0.2077 0.0556
0.0646 27.0 104301 0.2334 0.1997 0.0536
0.0607 28.0 108164 0.2530 0.2042 0.0554
0.0594 29.0 112027 0.2357 0.1957 0.0531
0.0553 30.0 115890 0.2497 0.1946 0.0529
0.0533 31.0 119753 0.2480 0.1944 0.0533
0.0495 32.0 123616 0.2506 0.1918 0.0524
0.0468 33.0 127479 0.2438 0.1821 0.0498
0.0445 34.0 131342 0.2578 0.1900 0.0528
0.0415 35.0 135205 0.2455 0.1787 0.0493
0.0393 36.0 139068 0.2615 0.1828 0.0509
0.0377 37.0 142931 0.2692 0.1721 0.0480
0.0346 38.0 146794 0.2640 0.1691 0.0475
0.0321 39.0 150657 0.2698 0.1666 0.0466
0.0312 40.0 154520 0.2790 0.1625 0.0457
0.0291 41.0 158383 0.2819 0.1634 0.0464
0.0276 42.0 162246 0.2794 0.1588 0.0449
0.0255 43.0 166109 0.2777 0.1568 0.0447
0.0241 44.0 169972 0.2910 0.1564 0.0446
0.0226 45.0 173835 0.2960 0.1543 0.0442
0.0209 46.0 177698 0.3081 0.1589 0.0462
0.0202 47.0 181561 0.2955 0.1511 0.0436
0.0192 48.0 185424 0.3025 0.1506 0.0433
0.0182 49.0 189287 0.3109 0.1502 0.0438
0.0165 50.0 193150 0.3240 0.1487 0.0434
0.0156 51.0 197013 0.3171 0.1465 0.0428
0.0151 52.0 200876 0.3327 0.1446 0.0424
0.014 53.0 204739 0.3338 0.1470 0.0431
0.0134 54.0 208602 0.3526 0.1431 0.0422
0.012 55.0 212465 0.3634 0.1397 0.0414
0.0118 56.0 216328 0.3425 0.1387 0.0412
0.011 57.0 220191 0.3449 0.1407 0.0417
0.0104 58.0 224054 0.3670 0.1370 0.0409
0.0099 59.0 227917 0.3684 0.1352 0.0402
0.0095 60.0 231780 0.3701 0.1363 0.0407
0.0088 61.0 235643 0.3810 0.1374 0.0409
0.008 62.0 239506 0.3758 0.1341 0.0402
0.008 63.0 243369 0.3826 0.1339 0.0401
0.0074 64.0 247232 0.3976 0.1356 0.0406
0.0069 65.0 251095 0.4055 0.1328 0.0398
0.0064 66.0 254958 0.3925 0.1324 0.0399
0.0059 67.0 258821 0.4302 0.1346 0.0408
0.0058 68.0 262684 0.4148 0.1272 0.0385
0.0051 69.0 266547 0.4322 0.1283 0.0389
0.0051 70.0 270410 0.4337 0.1278 0.0385
0.0046 71.0 274273 0.4233 0.1281 0.0389
0.0044 72.0 278136 0.4296 0.1282 0.0387
0.0039 73.0 281999 0.4433 0.1283 0.0389
0.0037 74.0 285862 0.4554 0.1245 0.0377
0.0033 75.0 289725 0.4776 0.1238 0.0377
0.0031 76.0 293588 0.4751 0.1246 0.0380
0.003 77.0 297451 0.4818 0.1238 0.0377
0.0027 78.0 301314 0.4832 0.1235 0.0377
0.0025 79.0 305177 0.5077 0.1218 0.0373
0.0022 80.0 309040 0.5017 0.1222 0.0375
0.0021 81.0 312903 0.5185 0.1239 0.0379
0.0019 82.0 316766 0.5278 0.1218 0.0374
0.0017 83.0 320629 0.5240 0.1221 0.0374
0.0016 84.0 324492 0.5542 0.1201 0.0369
0.0014 85.0 328355 0.5452 0.1198 0.0369
0.0013 86.0 332218 0.5652 0.1200 0.0369
0.0011 87.0 336081 0.5699 0.1186 0.0365
0.001 88.0 339944 0.5741 0.1190 0.0366
0.0009 89.0 343807 0.5721 0.1202 0.0370
0.0008 90.0 347670 0.6179 0.1177 0.0363
0.0006 91.0 351533 0.6347 0.1176 0.0364
0.0006 92.0 355396 0.6283 0.1164 0.0360
0.0005 93.0 359259 0.6473 0.1170 0.0362
0.0004 94.0 363122 0.6776 0.1162 0.0359
0.0003 95.0 366985 0.6908 0.1159 0.0358
0.0003 96.0 370848 0.7048 0.1157 0.0359
0.0002 97.0 374711 0.7340 0.1151 0.0357
0.0002 98.0 378574 0.7483 0.1150 0.0357
0.0001 99.0 382437 0.7624 0.1145 0.0356
0.0001 100.0 386300 0.7709 0.1144 0.0356

Framework versions

  • Transformers 4.51.3
  • Pytorch 2.6.0+cu126
  • Datasets 3.6.0
  • Tokenizers 0.21.1