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: 30.9
          - name: Wer
            type: wer
            value: 62.26924808644755

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: 1.3822
• Bleu: 30.9
• Chrf: 46.57
• Wer: 62.2692

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: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Bleu	Chrf	Validation Loss	Wer
2.3533	0.0438	100	6.29	25.08	1.7789	148.7618
1.9035	0.0876	200	18.21	34.02	1.5122	85.6821
1.5357	0.1313	300	14.01	33.7	1.3983	93.3363
1.3056	0.1751	400	18.12	37.35	1.3447	95.0023
1.1177	0.2189	500	18.47	38.44	1.3168	95.3624
0.984	0.2627	600	26.82	41.23	1.3202	67.3120
0.8945	0.3065	700	26.73	42.53	1.2947	67.1319
0.7508	0.3503	800	25.67	42.06	1.2476	74.2008
0.7127	0.3940	900	22.59	41.05	1.2630	75.7767
0.5944	0.4378	1000	22.37	40.31	1.2726	82.4854
0.4972	0.4816	1100	22.88	42.52	1.2898	82.5304
0.4517	0.5254	1200	27.99	44.42	1.2509	64.1603
0.3885	0.5692	1300	29.58	44.8	1.2887	63.1247
0.3337	0.6130	1400	30.05	45.5	1.2645	62.6294
0.2852	0.6567	1500	28.2	43.57	1.2972	68.6628
0.2583	0.7005	1600	28.21	45.06	1.2716	73.6155
0.2016	0.7443	1700	27.55	43.21	1.3346	74.3809
0.1883	0.7881	1800	21.45	41.83	1.3124	94.1018
0.1514	0.8319	1900	28.2	44.09	1.3178	63.7551
0.1311	0.8757	2000	27.33	43.25	1.3246	74.3359
0.1128	0.9194	2100	25.21	42.93	1.3464	83.2508
0.0994	0.9632	2200	30.51	45.74	1.3315	64.7456
0.0512	1.0070	2300	30.89	46.44	1.3377	63.3498
0.0447	1.0508	2400	28.72	44.36	1.3587	64.3404
0.0368	1.0946	2500	31.53	46.56	1.3619	61.9541
0.0281	1.1384	2600	30.98	46.45	1.3596	70.4638
0.0273	1.1821	2700	32.09	46.85	1.3656	62.1792
0.0287	1.2259	2800	32.57	47.04	1.3547	62.0891
0.025	1.2697	2900	26.94	45.43	1.3539	81.1796
0.0263	1.3135	3000	30.11	46.73	1.3512	71.4993
0.0301	1.3573	3100	1.3510	31.14	46.93	62.0891
0.0263	1.4011	3200	1.3853	31.64	46.98	61.6389
0.027	1.4448	3300	1.4148	29.63	45.91	65.1058
0.0286	1.4886	3400	1.3828	30.12	46.2	62.7195
0.0218	1.5324	3500	1.3890	30.41	46.28	64.8807
0.0231	1.5762	3600	1.3898	31.05	46.72	63.3498
0.0193	1.6200	3700	1.3836	30.05	45.7	62.4944
0.0184	1.6637	3800	1.3732	30.95	47.17	61.8640
0.0168	1.7075	3900	1.3780	30.9	46.91	62.1342
0.0168	1.7513	4000	1.3822	30.9	46.57	62.2692

Framework versions

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