README.md

metadata

language:
  - ga
  - en
license: apache-2.0
base_model: openai/whisper-medium
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
          type: ymoslem/IWSLT2023-GA-EN
        metrics:
          - name: Bleu
            type: bleu
            value: 33.1
          - name: Wer
            type: wer
            value: 62.40432237730752

Whisper Small GA-EN Speech Translation

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

• Loss: 1.1873
• Bleu: 33.1
• Chrf: 51.85
• Wer: 62.4043

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: 0.02
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Chrf	Wer
2.6291	0.0109	100	2.1971	2.33	16.34	175.5516
2.6591	0.0219	200	2.0357	5.57	22.49	122.2873
2.5637	0.0328	300	1.8690	7.67	26.29	133.0032
2.2954	0.0438	400	1.8062	11.2	30.03	114.2278
2.3292	0.0547	500	1.7421	9.85	29.28	117.2895
2.1223	0.0657	600	1.6739	14.56	32.56	84.2864
2.2398	0.0766	700	1.7187	13.86	34.74	98.9644
2.002	0.0876	800	1.6392	15.53	36.64	96.7582
1.8611	0.0985	900	1.6283	15.8	36.32	94.3719
1.8498	0.1095	1000	1.6102	17.58	36.0	85.5921
1.7585	0.1204	1100	1.6337	15.91	36.61	100.2251
1.6115	0.1314	1200	1.5381	22.21	39.94	76.8122
1.4415	0.1423	1300	1.5864	20.36	37.87	79.1986
1.5103	0.1533	1400	1.4925	23.2	41.26	75.2364
1.6576	0.1642	1500	1.4508	18.12	40.49	102.9266
1.3429	0.1752	1600	1.4399	27.88	43.74	69.7884
1.2522	0.1861	1700	1.4256	23.04	43.31	77.1724
1.2018	0.1970	1800	1.4072	21.06	40.39	78.6583
1.1945	0.2080	1900	1.4222	23.0	42.71	76.7222
1.1869	0.2189	2000	1.3992	22.54	42.02	75.8667
1.1752	0.2299	2100	1.3926	20.81	41.07	79.5137
1.0281	0.2408	2200	1.3633	27.24	45.55	69.6083
0.894	0.2518	2300	1.3287	28.6	45.58	65.8712
0.9788	0.2627	2400	1.3138	27.75	46.21	69.2931
0.8418	0.2737	2500	1.3064	27.85	46.17	68.3026
0.7559	0.2846	2600	1.2903	28.44	48.52	68.3476
0.8632	0.2956	2700	1.2834	27.87	46.86	68.3476
0.7501	0.3065	2800	1.2669	28.63	49.25	68.5277
0.6953	0.3175	2900	1.2615	30.46	48.83	64.4304
0.7195	0.3284	3000	1.2514	27.49	47.94	71.0941
0.6155	0.3394	3100	1.2428	30.06	49.64	66.5916
0.605	0.3503	3200	1.2040	31.64	50.27	63.8451
0.6349	0.3612	3300	1.2077	28.96	49.35	65.3760
0.4669	0.3722	3400	1.2219	31.17	48.95	64.2503
0.5196	0.3831	3500	1.2124	30.97	50.13	63.8001
0.5141	0.3941	3600	1.2026	31.97	50.8	63.0347
0.4221	0.4050	3700	1.1893	31.76	51.35	63.4399
0.2951	0.4160	3800	1.2049	32.4	51.08	63.1247
0.3898	0.4269	3900	1.1906	32.15	51.09	63.5299
0.4071	0.4379	4000	1.1873	33.1	51.85	62.4043

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.0+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1