README.md

metadata

library_name: transformers
language:
  - hi
license: apache-2.0
base_model: openai/whisper-medium
tags:
  - generated_from_trainer
datasets:
  - pranetk/paraspeak-data-new-stitched3
metrics:
  - wer
model-index:
  - name: Paraspeak Medium Test 6
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Paraspeak Hyperaugmented Dataset 2.0
          type: pranetk/paraspeak-data-new-stitched3
          args: 'config: hi, split: test'
        metrics:
          - name: Wer
            type: wer
            value: 0.055991041433370664

Paraspeak Medium Test 6

This model is a fine-tuned version of openai/whisper-medium on the Paraspeak Hyperaugmented Dataset 2.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0008
• Wer: 0.0560

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: 1
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 10
• num_epochs: 7
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3433	0.1721	100	0.2304	32.6708
0.0793	0.3442	200	0.1057	16.1814
0.0786	0.5164	300	0.0822	18.7850
0.0473	0.6885	400	0.0412	7.5868
0.0518	0.8606	500	0.0374	5.9910
0.0249	1.0327	600	0.0305	5.0392
0.0638	1.2048	700	0.0279	6.0470
0.013	1.3769	800	0.0266	3.6394
0.0117	1.5491	900	0.0169	3.1355
0.0011	1.7212	1000	0.0101	2.2116
0.0058	1.8933	1100	0.0112	1.5957
0.0021	2.0654	1200	0.0119	1.4558
0.0014	2.2375	1300	0.0086	1.4278
0.0062	2.4096	1400	0.0103	1.3158
0.0063	2.5818	1500	0.0135	2.0157
0.013	2.7539	1600	0.0116	1.7637
0.0019	2.9260	1700	0.0068	0.8399
0.0131	3.0981	1800	0.0118	0.6719
0.002	3.2702	1900	0.0073	0.7279
0.0002	3.4423	2000	0.0057	0.5039
0.0001	3.6145	2100	0.0036	0.4199
0.0226	3.7866	2200	0.0066	0.5319
0.0038	3.9587	2300	0.0020	0.1400
0.0297	4.1308	2400	0.0045	0.3080
0.0	4.3029	2500	0.0026	0.0840
0.0	4.4750	2600	0.0040	0.3359
0.0	4.6472	2700	0.0006	0.0560
0.0	4.8193	2800	0.0009	0.0840
0.0	4.9914	2900	0.0010	0.0840
0.0	5.1635	3000	0.0021	0.0560
0.0	5.3356	3100	0.0021	0.2520
0.0102	5.5077	3200	0.0020	0.1960
0.0	5.6799	3300	0.0013	0.0560
0.0	5.8520	3400	0.0011	0.0560
0.0	6.0241	3500	0.0010	0.0560
0.0	6.1962	3600	0.0011	0.0560
0.0	6.3683	3700	0.0012	0.0560
0.0	6.5404	3800	0.0012	0.0560
0.0	6.7126	3900	0.0008	0.0560
0.0	6.8847	4000	0.0008	0.0560

Framework versions

• Transformers 4.46.2
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3