miosipof/asr_Synth_EasyCall_v0.2

Whisper Medium

This model is a fine-tuned version of openai/whisper-medium on the Dysarthria_Synthetic_Easycall_Common dataset. It achieves the following results on the evaluation set:

• Loss: 0.9647
• Wer: 70.3226

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 16
• 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: reduce_lr_on_plateau
• lr_scheduler_warmup_steps: 100
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.0651	0.6897	50	5.7375	162.9032
4.9066	1.3793	100	5.3309	93.5484
4.2186	2.0690	150	4.7377	89.3548
3.628	2.7586	200	4.0118	81.6129
3.0181	3.4483	250	3.6058	81.6129
2.6682	4.1379	300	3.2728	78.3871
2.3183	4.8276	350	2.8405	75.4839
1.9084	5.5172	400	2.1841	71.9355
1.2191	6.2069	450	1.2534	73.5484
0.8483	6.8966	500	1.1853	73.5484
0.8115	7.5862	550	1.1503	70.9677
0.7512	8.2759	600	1.1072	73.5484
0.7064	8.9655	650	1.0806	74.5161
0.6779	9.6552	700	1.0477	75.1613
0.6345	10.3448	750	1.0352	74.8387
0.6097	11.0345	800	1.0186	73.8710
0.5927	11.7241	850	1.0120	74.5161
0.5619	12.4138	900	1.0017	73.5484
0.5349	13.1034	950	0.9879	70.0
0.5207	13.7931	1000	0.9900	70.0
0.5168	14.4828	1050	0.9828	173.8710
0.4703	15.1724	1100	0.9676	73.5484
0.476	15.8621	1150	0.9729	175.8065
0.4443	16.5517	1200	0.9652	74.1935
0.4215	17.2414	1250	0.9635	176.1290
0.4206	17.9310	1300	0.9631	179.3548
0.3971	18.6207	1350	0.9687	70.6452
0.3838	19.3103	1400	0.9543	147.7419
0.3791	20.0	1450	0.9594	68.3871
0.3441	20.6897	1500	0.9608	257.0968
0.3574	21.3793	1550	0.9589	71.2903
0.3323	22.0690	1600	0.9619	69.6774
0.3187	22.7586	1650	0.9552	154.1935
0.3011	23.4483	1700	0.9568	67.0968
0.2997	24.1379	1750	0.9580	70.9677
0.287	24.8276	1800	0.9566	70.6452
0.2747	25.5172	1850	0.9694	68.3871
0.2665	26.2069	1900	0.9544	69.0323
0.2565	26.8966	1950	0.9518	155.8065
0.2448	27.5862	2000	0.9579	68.7097
0.2377	28.2759	2050	0.9589	67.0968
0.2323	28.9655	2100	0.9590	68.7097
0.2224	29.6552	2150	0.9647	70.3226

Framework versions

• PEFT 0.14.0
• Transformers 4.45.2
• Pytorch 2.2.0
• Datasets 3.2.0
• Tokenizers 0.20.3