miosipof/asr_Synth_EasyCall_v0.1

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: 0.9709
• Wer: 66.7742

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: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.1405	0.6897	50	5.7981	205.1613
4.968	1.3793	100	5.3200	164.8387
4.2444	2.0690	150	4.7663	85.8065
3.6175	2.7586	200	4.0413	83.8710
3.0351	3.4483	250	3.6257	80.3226
2.668	4.1379	300	3.2450	75.8065
2.3178	4.8276	350	2.8196	75.1613
1.9613	5.5172	400	2.3047	73.8710
1.5398	6.2069	450	1.8564	73.8710
1.2771	6.8966	500	1.5804	76.4516
1.0752	7.5862	550	1.3762	75.1613
0.8899	8.2759	600	1.2049	77.0968
0.7684	8.9655	650	1.1252	75.1613
0.7242	9.6552	700	1.0776	74.1935
0.67	10.3448	750	1.0567	72.9032
0.6405	11.0345	800	1.0334	73.2258
0.622	11.7241	850	1.0179	74.1935
0.5874	12.4138	900	1.0067	71.2903
0.5601	13.1034	950	0.9863	67.0968
0.544	13.7931	1000	0.9855	67.0968
0.538	14.4828	1050	0.9709	66.7742

Framework versions

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