ethiopic-asr

This model is a fine-tuned version of Samuael/geez-asr on the alffa_amharic dataset. It achieves the following results on the evaluation set:

• Loss: 0.1301
• Wer: 0.1469
• Phoneme Cer: 0.0296
• Cer: 0.0416

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: 3e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 1
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Phoneme Cer	Cer
No log	0.0442	200	3.2216	1.0	1.0	1.0
No log	0.0883	400	3.1164	1.0	1.0	1.0
4.1769	0.1325	600	0.9628	0.5476	0.1141	0.1609
4.1769	0.1767	800	0.3181	0.2150	0.0430	0.0607
0.8455	0.2208	1000	0.2195	0.1759	0.0353	0.0503
0.8455	0.2650	1200	0.1913	0.1846	0.0365	0.0520
0.8455	0.3092	1400	0.1699	0.1591	0.0322	0.0454
0.2929	0.3534	1600	0.1603	0.1572	0.0316	0.0442
0.2929	0.3975	1800	0.1503	0.1567	0.0315	0.0442
0.2392	0.4417	2000	0.1476	0.1587	0.0318	0.0446
0.2392	0.4859	2200	0.1449	0.1565	0.0312	0.0438
0.2392	0.5300	2400	0.1409	0.1537	0.0308	0.0427
0.2166	0.5742	2600	0.1395	0.1551	0.0308	0.0428
0.2166	0.6184	2800	0.1345	0.1469	0.0290	0.0410
0.2068	0.6625	3000	0.1331	0.1509	0.0297	0.0419
0.2068	0.7067	3200	0.1346	0.1518	0.0301	0.0421
0.2068	0.7509	3400	0.1335	0.1507	0.0303	0.0426
0.2037	0.7951	3600	0.1312	0.1471	0.0297	0.0415
0.2037	0.8392	3800	0.1303	0.1438	0.0289	0.0406
0.1985	0.8834	4000	0.1300	0.1457	0.0292	0.0410
0.1985	0.9276	4200	0.1303	0.1471	0.0295	0.0414
0.1985	0.9717	4400	0.1301	0.1469	0.0296	0.0416

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0