Whisper Tiny es

This model is a fine-tuned version of openai/whisper-tiny on the following datasets:

deepdml/common_voice_26_0
disco-eth/WorldSpeech
facebook/voxpopuli
facebook/multilingual_librispeech
deepdml/voxforge
google/fleurs
deepdml/basque_parliament_1

It achieves the following results on the evaluation set:

Best checkpoint: 44,000
Loss: 0.2888
Wer: 16.0693
Cer: 6.0413

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: 1e-05
train_batch_size: 128
eval_batch_size: 128
seed: 42
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: 0.04
training_steps: 46000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Raw	Cer Raw	Wer	Cer
0.3663	0.0217	1000	0.5971	31.8722	12.1478	31.6510	12.1036
0.2625	0.0435	2000	0.5140	28.2789	10.5567	28.2345	10.5484
0.2183	0.0652	3000	0.4763	25.3893	9.2745	25.3683	9.2709
0.2320	0.0870	4000	0.4435	24.3033	9.0073	24.2944	9.0057
0.3290	0.1087	5000	0.4363	23.9673	9.1117	23.9590	9.1103
0.2030	0.1304	6000	0.4118	22.9428	8.5741	22.9390	8.5735
0.3456	0.1522	7000	0.4131	22.7190	8.6583	22.7139	8.6574
0.3284	0.1739	8000	0.3996	22.6752	8.4567	22.6727	8.4563
0.2336	0.1957	9000	0.3692	20.9236	7.9288	20.9223	7.9286
0.2950	0.2174	10000	0.3622	21.1436	8.2788	21.1436	8.2788
0.3875	0.2391	11000	0.3575	20.1197	7.5460	20.1197	7.5460
0.1884	1.0020	12000	0.3406	19.3906	7.2308	19.3906	7.2308
0.1370	1.0237	13000	0.3373	19.3215	7.3625	19.3215	7.3625
0.1426	1.0455	14000	0.3364	19.1237	7.1903	19.1237	7.1903
0.1616	1.0672	15000	0.3329	18.5785	6.9592	18.5785	6.9592
0.1560	1.0889	16000	0.3278	19.0952	7.3977	19.0952	7.3977
0.1511	1.1107	17000	0.3308	18.7877	7.2803	18.7877	7.2803
0.1642	1.1324	18000	0.3276	18.3573	6.7823	18.3573	6.7823
0.2495	1.1542	19000	0.3300	18.6108	7.0579	18.6108	7.0579
0.1903	1.1759	20000	0.3232	18.4504	7.0561	18.4504	7.0561
0.1982	1.1976	21000	0.3161	18.3636	7.0710	18.3636	7.0710
0.3891	1.2194	22000	0.3154	18.5240	7.1787	18.5240	7.1787
0.2725	1.2411	23000	0.3155	17.8260	6.6920	17.8260	6.6920
0.1271	2.0040	24000	0.3016	17.1540	6.5646	17.1540	6.5646
0.1171	2.0257	25000	0.3090	17.2700	6.5550	17.2700	6.5550
0.1738	2.0474	26000	0.3037	17.3721	6.6015	17.3721	6.6015
0.1532	2.0692	27000	0.3069	17.1977	6.4330	17.1977	6.4330
0.2220	2.0909	28000	0.3056	17.4139	6.5979	17.4139	6.5979
0.1509	2.1127	29000	0.3010	16.9923	6.5297	16.9923	6.5297
0.1484	2.1344	30000	0.3033	16.7863	6.2084	16.7863	6.2084
0.1558	2.1561	31000	0.3021	17.1407	6.5016	17.1407	6.5016
0.1563	2.1779	32000	0.2979	16.6677	6.2236	16.6677	6.2236
0.2823	2.1996	33000	0.3006	17.1850	6.4912	17.1850	6.4912
0.1463	2.2213	34000	0.2955	16.6018	6.3068	16.6018	6.3068
0.1209	2.2431	35000	0.2916	16.6373	6.3950	16.6373	6.3950
0.1145	3.0059	36000	0.2941	16.8801	6.5048	16.8801	6.5048
0.1545	3.0277	37000	0.2925	16.6208	6.4095	16.6208	6.4095
0.3211	3.0494	38000	0.2905	16.6354	6.4966	16.6354	6.4966
0.1914	3.0712	39000	0.2926	16.2645	6.1487	16.2645	6.1487
0.1672	3.0929	40000	0.2936	16.5517	6.2828	16.5517	6.2828
0.1256	3.1146	41000	0.2910	16.6861	6.5220	16.6861	6.5220
0.1023	3.1364	42000	0.2918	16.4179	6.2376	16.4179	6.2376
0.1597	3.1581	43000	0.2913	16.6151	6.4257	16.6151	6.4257
0.1912	3.1798	44000	0.2888	16.0693	6.0413	16.0693	6.0413
0.1459	3.2016	45000	0.2891	16.2379	6.1107	16.2379	6.1107
0.2485	3.2233	46000	0.2889	16.3450	6.1852	16.3450	6.1852

Framework versions

Transformers 5.14.1
Pytorch 2.6.0+cu124
Datasets 5.0.1
Tokenizers 0.22.2

Citation

Please cite the model using the following BibTeX entry:

@misc{deepdml/whisper-tiny-es-mix-norm,
      title={Fine-tuned Whisper tiny ASR model for speech recognition in Spanish},
      author={Jimenez, David},
      howpublished={\url{https://huggingface.co/deepdml/whisper-tiny-es-mix-norm}},
      year={2026}
    }