wav2vec2-large-xls-r-300m-Arabic-phoneme

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

• Loss: 0.0335
• Per: 0.0199
• Wer: 0.0225

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: 0.0005
• train_batch_size: 8
• eval_batch_size: 6
• 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: 250
• num_epochs: 30.0

Training results

Training Loss	Epoch	Step	Validation Loss	Per	Wer
5.3718	1.0	102	2.1140	1.0	1.0
2.036	2.0	204	2.0637	1.0	1.0
2.0175	3.0	306	2.1252	1.0	1.0
1.9463	4.0	408	1.7014	0.9942	0.9887
1.702	5.0	510	1.7257	0.9944	0.9892
1.6475	6.0	612	1.5855	0.9897	0.9798
1.4766	7.0	714	1.2777	0.9787	0.9641
1.0363	8.0	816	0.7926	0.7738	0.7731
0.5964	9.0	918	0.4220	0.3994	0.4133
0.3437	10.0	1020	0.2307	0.1387	0.1549
0.2052	11.0	1122	0.1587	0.0645	0.0738
0.1509	12.0	1224	0.1314	0.0464	0.0544
0.1256	13.0	1326	0.1070	0.0448	0.0518
0.0935	14.0	1428	0.0854	0.0394	0.0452
0.0779	15.0	1530	0.0896	0.0376	0.0440
0.0674	16.0	1632	0.0625	0.0255	0.0306
0.0558	17.0	1734	0.0573	0.0270	0.0318
0.0492	18.0	1836	0.0542	0.0248	0.0288
0.0486	19.0	1938	0.0631	0.0336	0.0369
0.047	20.0	2040	0.0482	0.0255	0.0290
0.0432	21.0	2142	0.0470	0.0262	0.0307
0.0433	22.0	2244	0.0460	0.0250	0.0290
0.0367	23.0	2346	0.0450	0.0253	0.0295
0.0343	24.0	2448	0.0444	0.0254	0.0283
0.0292	25.0	2550	0.0427	0.0248	0.0283
0.0261	26.0	2652	0.0409	0.0220	0.0250
0.025	27.0	2754	0.0360	0.0221	0.0251
0.0236	28.0	2856	0.0350	0.0208	0.0231
0.0222	29.0	2958	0.0338	0.0199	0.0222
0.0202	30.0	3060	0.0335	0.0199	0.0225

Framework versions

• Transformers 4.33.1
• Pytorch 2.0.1+cu118
• Datasets 1.18.3
• Tokenizers 0.13.3