README.md

metadata

language:
  - ga
  - en
license: apache-2.0
base_model: openai/whisper-small
tags:
  - generated_from_trainer
datasets:
  - ymoslem/IWSLT2023-GA-EN
  - ymoslem/FLEURS-GA-EN
  - ymoslem/BitesizeIrish-GA-EN
  - ymoslem/SpokenWords-GA-EN-MTed
  - ymoslem/Tatoeba-Speech-Irish
  - ymoslem/Wikimedia-Speech-Irish
metrics:
  - bleu
  - wer
model-index:
  - name: Whisper Small GA-EN Speech Translation
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: >-
            IWSLT-2023, FLEURS, BiteSize, SpokenWords, Tatoeba, and Wikimedia +
            augmented
          type: ymoslem/IWSLT2023-GA-EN
        metrics:
          - name: Bleu
            type: bleu
            value: 28.44
          - name: Wer
            type: wer
            value: 72.62494371904548

Whisper Small GA-EN Speech Translation

This model is a fine-tuned version of openai/whisper-small on the IWSLT-2023, FLEURS, BiteSize, SpokenWords, Tatoeba, and Wikimedia + augmented dataset. It achieves the following results on the evaluation set:

• Loss: 1.3641
• Bleu: 28.44
• Chrf: 43.55
• Wer: 72.6249

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: 0.0001
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Chrf	Wer
2.3595	0.0438	100	1.7944	9.69	26.37	114.4529
1.9008	0.0876	200	1.5391	14.89	32.44	93.6065
1.535	0.1313	300	1.3972	18.24	33.57	81.9901
1.3307	0.1751	400	1.3684	21.34	37.37	72.8050
1.1263	0.2189	500	1.3284	19.33	39.83	91.8955
0.9805	0.2627	600	1.3301	23.67	38.68	78.3881
0.8989	0.3065	700	1.3123	20.32	36.94	76.3170
0.7557	0.3503	800	1.2717	25.74	40.16	72.4448
0.7216	0.3940	900	1.3090	22.34	37.79	78.9284
0.6131	0.4378	1000	1.2566	24.36	41.49	74.5160
0.5032	0.4816	1100	1.2742	21.69	41.12	83.3859
0.4567	0.5254	1200	1.2893	24.33	40.05	70.8690
0.3968	0.5692	1300	1.3000	26.97	41.45	69.6083
0.3353	0.6130	1400	1.2784	27.51	43.97	63.9352
0.2826	0.6567	1500	1.3165	24.36	39.83	70.6439
0.2643	0.7005	1600	1.3317	24.98	40.01	68.6628
0.2047	0.7443	1700	1.2905	28.01	42.72	65.8262
0.1946	0.7881	1800	1.2820	26.17	42.46	64.9257
0.1588	0.8319	1900	1.3172	26.9	43.02	63.5299
0.1322	0.8757	2000	1.3248	27.78	43.53	63.8001
0.1134	0.9194	2100	1.3198	28.98	45.27	72.7600
0.1031	0.9632	2200	1.3502	29.18	44.77	68.3476
0.0518	1.0070	2300	1.3433	28.6	42.96	69.0230
0.0481	1.0508	2400	1.3715	29.01	44.46	69.6983
0.0367	1.0946	2500	1.3696	26.94	42.39	73.6605
0.0309	1.1384	2600	1.3665	28.12	43.32	70.3737
0.0302	1.1821	2700	1.3836	29.6	44.56	67.2220
0.0302	1.2259	2800	1.3667	29.0	44.33	67.2220
0.0252	1.2697	2900	1.3633	29.07	44.09	70.6889
0.0257	1.3135	3000	1.3641	28.44	43.55	72.6249

Framework versions

• Transformers 4.40.2
• Pytorch 2.2.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1