xls-r-es-test-lm
This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - ES dataset. It achieves the following results on the test set with lm model:
- Loss: 0.1304
- WER: 0.094
- CER: 0.031
It achieves the following results on the val set with lm model:
- Loss: 0.1304
- WER: 0.081
- CER: 0.025
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: 7.5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- 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: 10.0
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
2.9613 | 0.07 | 500 | 2.9647 | 1.0 |
2.604 | 0.14 | 1000 | 1.8300 | 0.9562 |
1.177 | 0.21 | 1500 | 0.3652 | 0.3077 |
1.0745 | 0.28 | 2000 | 0.2707 | 0.2504 |
1.0103 | 0.35 | 2500 | 0.2338 | 0.2157 |
0.9858 | 0.42 | 3000 | 0.2321 | 0.2129 |
0.974 | 0.49 | 3500 | 0.2164 | 0.2031 |
0.9699 | 0.56 | 4000 | 0.2078 | 0.1970 |
0.9513 | 0.63 | 4500 | 0.2173 | 0.2139 |
0.9657 | 0.7 | 5000 | 0.2050 | 0.1979 |
0.9484 | 0.77 | 5500 | 0.2008 | 0.1919 |
0.9317 | 0.84 | 6000 | 0.2012 | 0.1911 |
0.9366 | 0.91 | 6500 | 0.2024 | 0.1976 |
0.9242 | 0.98 | 7000 | 0.2062 | 0.2028 |
0.9138 | 1.05 | 7500 | 0.1924 | 0.1863 |
0.921 | 1.12 | 8000 | 0.1935 | 0.1836 |
0.9117 | 1.19 | 8500 | 0.1887 | 0.1815 |
0.9064 | 1.26 | 9000 | 0.1909 | 0.1839 |
0.9118 | 1.32 | 9500 | 0.1869 | 0.1830 |
0.9121 | 1.39 | 10000 | 0.1863 | 0.1802 |
0.9048 | 1.46 | 10500 | 0.1845 | 0.1791 |
0.8955 | 1.53 | 11000 | 0.1863 | 0.1774 |
0.8947 | 1.6 | 11500 | 0.1907 | 0.1814 |
0.9073 | 1.67 | 12000 | 0.1892 | 0.1853 |
0.8927 | 1.74 | 12500 | 0.1821 | 0.1750 |
0.8732 | 1.81 | 13000 | 0.1815 | 0.1768 |
0.8761 | 1.88 | 13500 | 0.1822 | 0.1749 |
0.8751 | 1.95 | 14000 | 0.1789 | 0.1715 |
0.8889 | 2.02 | 14500 | 0.1819 | 0.1791 |
0.8864 | 2.09 | 15000 | 0.1826 | 0.1794 |
0.886 | 2.16 | 15500 | 0.1788 | 0.1776 |
0.8915 | 2.23 | 16000 | 0.1756 | 0.1719 |
0.8689 | 2.3 | 16500 | 0.1769 | 0.1711 |
0.879 | 2.37 | 17000 | 0.1777 | 0.1739 |
0.8692 | 2.44 | 17500 | 0.1765 | 0.1705 |
0.8504 | 2.51 | 18000 | 0.1699 | 0.1652 |
0.8728 | 2.58 | 18500 | 0.1705 | 0.1694 |
0.8523 | 2.65 | 19000 | 0.1674 | 0.1645 |
0.8513 | 2.72 | 19500 | 0.1661 | 0.1611 |
0.8498 | 2.79 | 20000 | 0.1660 | 0.1631 |
0.8432 | 2.86 | 20500 | 0.1636 | 0.1610 |
0.8492 | 2.93 | 21000 | 0.1708 | 0.1688 |
0.8561 | 3.0 | 21500 | 0.1663 | 0.1604 |
0.842 | 3.07 | 22000 | 0.1690 | 0.1625 |
0.857 | 3.14 | 22500 | 0.1642 | 0.1605 |
0.8518 | 3.21 | 23000 | 0.1626 | 0.1585 |
0.8506 | 3.28 | 23500 | 0.1651 | 0.1605 |
0.8394 | 3.35 | 24000 | 0.1647 | 0.1585 |
0.8431 | 3.42 | 24500 | 0.1632 | 0.1573 |
0.8566 | 3.49 | 25000 | 0.1614 | 0.1550 |
0.8534 | 3.56 | 25500 | 0.1645 | 0.1589 |
0.8386 | 3.63 | 26000 | 0.1632 | 0.1582 |
0.8357 | 3.7 | 26500 | 0.1631 | 0.1556 |
0.8299 | 3.77 | 27000 | 0.1612 | 0.1550 |
0.8421 | 3.84 | 27500 | 0.1602 | 0.1552 |
0.8375 | 3.91 | 28000 | 0.1592 | 0.1537 |
0.8328 | 3.97 | 28500 | 0.1587 | 0.1537 |
0.8155 | 4.04 | 29000 | 0.1587 | 0.1520 |
0.8335 | 4.11 | 29500 | 0.1624 | 0.1556 |
0.8138 | 4.18 | 30000 | 0.1581 | 0.1547 |
0.8195 | 4.25 | 30500 | 0.1560 | 0.1507 |
0.8092 | 4.32 | 31000 | 0.1561 | 0.1534 |
0.8191 | 4.39 | 31500 | 0.1549 | 0.1493 |
0.8008 | 4.46 | 32000 | 0.1540 | 0.1493 |
0.8138 | 4.53 | 32500 | 0.1544 | 0.1493 |
0.8173 | 4.6 | 33000 | 0.1553 | 0.1511 |
0.8081 | 4.67 | 33500 | 0.1541 | 0.1484 |
0.8192 | 4.74 | 34000 | 0.1560 | 0.1506 |
0.8068 | 4.81 | 34500 | 0.1540 | 0.1503 |
0.8105 | 4.88 | 35000 | 0.1529 | 0.1483 |
0.7976 | 4.95 | 35500 | 0.1507 | 0.1451 |
0.8143 | 5.02 | 36000 | 0.1505 | 0.1462 |
0.8053 | 5.09 | 36500 | 0.1517 | 0.1476 |
0.785 | 5.16 | 37000 | 0.1526 | 0.1478 |
0.7936 | 5.23 | 37500 | 0.1489 | 0.1421 |
0.807 | 5.3 | 38000 | 0.1483 | 0.1420 |
0.8092 | 5.37 | 38500 | 0.1481 | 0.1435 |
0.793 | 5.44 | 39000 | 0.1503 | 0.1438 |
0.814 | 5.51 | 39500 | 0.1495 | 0.1480 |
0.807 | 5.58 | 40000 | 0.1472 | 0.1424 |
0.7913 | 5.65 | 40500 | 0.1471 | 0.1422 |
0.7844 | 5.72 | 41000 | 0.1473 | 0.1422 |
0.7888 | 5.79 | 41500 | 0.1445 | 0.1385 |
0.7806 | 5.86 | 42000 | 0.1435 | 0.1394 |
0.7773 | 5.93 | 42500 | 0.1461 | 0.1424 |
0.786 | 6.0 | 43000 | 0.1450 | 0.1413 |
0.7784 | 6.07 | 43500 | 0.1463 | 0.1424 |
0.7937 | 6.14 | 44000 | 0.1438 | 0.1386 |
0.7738 | 6.21 | 44500 | 0.1437 | 0.1383 |
0.7728 | 6.28 | 45000 | 0.1424 | 0.1371 |
0.7681 | 6.35 | 45500 | 0.1416 | 0.1376 |
0.776 | 6.42 | 46000 | 0.1415 | 0.1380 |
0.7773 | 6.49 | 46500 | 0.1416 | 0.1371 |
0.7692 | 6.56 | 47000 | 0.1398 | 0.1345 |
0.7642 | 6.62 | 47500 | 0.1381 | 0.1341 |
0.7692 | 6.69 | 48000 | 0.1392 | 0.1334 |
0.7667 | 6.76 | 48500 | 0.1392 | 0.1348 |
0.7712 | 6.83 | 49000 | 0.1398 | 0.1333 |
0.7628 | 6.9 | 49500 | 0.1392 | 0.1344 |
0.7622 | 6.97 | 50000 | 0.1377 | 0.1329 |
0.7639 | 7.04 | 50500 | 0.1361 | 0.1316 |
0.742 | 7.11 | 51000 | 0.1376 | 0.1327 |
0.7526 | 7.18 | 51500 | 0.1387 | 0.1342 |
0.7606 | 7.25 | 52000 | 0.1363 | 0.1316 |
0.7626 | 7.32 | 52500 | 0.1365 | 0.1313 |
0.752 | 7.39 | 53000 | 0.1354 | 0.1309 |
0.7562 | 7.46 | 53500 | 0.1362 | 0.1312 |
0.7557 | 7.53 | 54000 | 0.1358 | 0.1325 |
0.7588 | 7.6 | 54500 | 0.1343 | 0.1311 |
0.7485 | 7.67 | 55000 | 0.1346 | 0.1301 |
0.7466 | 7.74 | 55500 | 0.1354 | 0.1314 |
0.7558 | 7.81 | 56000 | 0.1359 | 0.1325 |
0.7578 | 7.88 | 56500 | 0.1363 | 0.1334 |
0.7411 | 7.95 | 57000 | 0.1346 | 0.1301 |
0.7478 | 8.02 | 57500 | 0.1355 | 0.1305 |
0.7451 | 8.09 | 58000 | 0.1349 | 0.1302 |
0.7383 | 8.16 | 58500 | 0.1349 | 0.1294 |
0.7482 | 8.23 | 59000 | 0.1341 | 0.1293 |
0.742 | 8.3 | 59500 | 0.1338 | 0.1296 |
0.7343 | 8.37 | 60000 | 0.1348 | 0.1307 |
0.7385 | 8.44 | 60500 | 0.1324 | 0.1282 |
0.7567 | 8.51 | 61000 | 0.1334 | 0.1281 |
0.7342 | 8.58 | 61500 | 0.1338 | 0.1289 |
0.7401 | 8.65 | 62000 | 0.1331 | 0.1285 |
0.7362 | 8.72 | 62500 | 0.1329 | 0.1283 |
0.7241 | 8.79 | 63000 | 0.1323 | 0.1277 |
0.7244 | 8.86 | 63500 | 0.1317 | 0.1269 |
0.7274 | 8.93 | 64000 | 0.1308 | 0.1260 |
0.7411 | 9.0 | 64500 | 0.1309 | 0.1256 |
0.7255 | 9.07 | 65000 | 0.1316 | 0.1265 |
0.7406 | 9.14 | 65500 | 0.1315 | 0.1270 |
0.7418 | 9.21 | 66000 | 0.1315 | 0.1269 |
0.7301 | 9.27 | 66500 | 0.1315 | 0.1273 |
0.7248 | 9.34 | 67000 | 0.1323 | 0.1274 |
0.7423 | 9.41 | 67500 | 0.1309 | 0.1267 |
0.7152 | 9.48 | 68000 | 0.1312 | 0.1271 |
0.7295 | 9.55 | 68500 | 0.1306 | 0.1262 |
0.7231 | 9.62 | 69000 | 0.1308 | 0.1263 |
0.7344 | 9.69 | 69500 | 0.1313 | 0.1267 |
0.7264 | 9.76 | 70000 | 0.1305 | 0.1263 |
0.7309 | 9.83 | 70500 | 0.1303 | 0.1262 |
0.73 | 9.9 | 71000 | 0.1303 | 0.1261 |
0.7353 | 9.97 | 71500 | 0.1304 | 0.1260 |
Framework versions
- Transformers 4.17.0.dev0
- Pytorch 1.10.2+cu102
- Datasets 1.18.3
- Tokenizers 0.11.0
- Downloads last month
- 15
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 glob-asr/xls-r-es-test-lm
Space using glob-asr/xls-r-es-test-lm 1
Evaluation results
- Test WER on Common Voice 8.0self-reported9.400
- Test WER on Robust Speech Event - Dev Dataself-reported27.950
- Test WER on Robust Speech Event - Test Dataself-reported30.860