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 Medium GA-EN Speech Translation
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: IWSLT-2023, FLEURS, BiteSize, SpokenWords, Tatoeba, and Wikimedia
          type: ymoslem/IWSLT2023-GA-EN
        metrics:
          - name: Bleu
            type: bleu
            value: 32.53
          - name: Wer
            type: wer
            value: 62.809545249887435

Whisper Medium 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 dataset. It achieves the following results on the evaluation set:

• Loss: 1.2087
• Bleu: 32.53
• Chrf: 52.88
• Wer: 62.8095

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.03
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Chrf	Wer
2.5164	0.0328	100	2.0060	2.56	17.46	162.9896
2.656	0.0657	200	2.0232	8.49	26.0	99.5498
2.5156	0.0985	300	1.9253	7.55	25.1	141.2877
2.4722	0.1314	400	1.8289	12.52	30.49	90.4548
2.3376	0.1642	500	1.6839	17.39	33.23	81.1796
2.1733	0.1970	600	1.7342	9.62	32.48	137.9559
2.3382	0.2299	700	1.6570	12.54	34.43	112.2467
2.0041	0.2627	800	1.6048	17.55	36.73	85.1418
2.1142	0.2956	900	1.6256	17.58	35.74	82.7105
2.024	0.3284	1000	1.5861	14.4	37.22	86.7177
1.7556	0.3612	1100	1.5415	17.21	38.88	84.5115
1.6904	0.3941	1200	1.4902	19.6	38.84	85.3670
1.674	0.4269	1300	1.4748	20.33	41.3	88.3836
1.6899	0.4598	1400	1.4479	22.74	43.25	80.9995
1.5234	0.4926	1500	1.3763	20.13	42.08	80.6844
1.364	0.5255	1600	1.4164	23.12	41.78	72.9851
1.5267	0.5583	1700	1.3855	19.94	41.63	91.7605
1.4282	0.5911	1800	1.3729	23.96	44.84	74.6961
1.3611	0.6240	1900	1.3562	23.1	45.41	81.8100
1.1396	0.6568	2000	1.3131	27.9	46.89	67.2670
1.1849	0.6897	2100	1.3483	24.38	45.25	75.8667
1.0871	0.7225	2200	1.2848	28.64	48.93	66.6817
1.1822	0.7553	2300	1.2782	28.41	47.25	68.6628
1.1272	0.7882	2400	1.2549	27.24	48.57	75.9568
1.0241	0.8210	2500	1.2922	25.74	47.44	74.4710
0.9629	0.8539	2600	1.3209	23.93	44.61	82.1252
0.8251	0.8867	2700	1.2273	32.21	51.64	65.5110
0.7921	0.9195	2800	1.2881	26.38	48.31	80.2792
0.8873	0.9524	2900	1.2268	26.57	50.09	77.1724
0.7967	0.9852	3000	1.2036	29.35	51.53	69.6533
0.3119	1.0181	3100	1.2231	31.77	51.57	62.3143
0.3009	1.0509	3200	1.2446	31.8	50.44	61.8190
0.2855	1.0837	3300	1.2240	30.48	50.86	66.7717
0.2535	1.1166	3400	1.2287	31.96	52.82	63.3949
0.2162	1.1494	3500	1.2398	33.91	52.17	61.3688
0.2307	1.1823	3600	1.2280	32.11	51.67	64.7456
0.2184	1.2151	3700	1.2149	34.59	53.32	59.9730
0.2365	1.2479	3800	1.2044	32.51	52.98	62.3593
0.1958	1.2808	3900	1.2116	32.45	52.86	63.1697
0.2081	1.3136	4000	1.2087	32.53	52.88	62.8095

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1