whisper-small-noisy-hindi-10dB

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.7442
Wer: 41.8554

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: 64
eval_batch_size: 32
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
training_steps: 3000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.6146	0.61	50	1.3244	85.2585
0.8209	1.22	100	0.7607	55.4556
0.6434	1.83	150	0.6184	51.5822
0.5053	2.44	200	0.5191	46.7404
0.409	3.05	250	0.4271	41.9938
0.265	3.66	300	0.3151	39.4778
0.1786	4.27	350	0.2965	37.3076
0.1617	4.88	400	0.2826	36.2355
0.103	5.49	450	0.2877	35.5957
0.0907	6.1	500	0.2929	35.3450
0.0595	6.71	550	0.3032	34.8262
0.0338	7.32	600	0.3186	34.7743
0.0365	7.93	650	0.3303	34.3853
0.021	8.54	700	0.3414	34.3420
0.0174	9.15	750	0.3561	34.1605
0.0129	9.76	800	0.3619	34.3247
0.009	10.37	850	0.3681	33.9703
0.0082	10.98	900	0.3802	34.2469
0.006	11.59	950	0.3817	33.4083
0.0052	12.2	1000	0.4054	34.4112
0.005	12.8	1050	0.4113	34.2123
0.0041	13.41	1100	0.4139	33.8060
0.0043	14.02	1150	0.4161	32.9500
0.0028	14.63	1200	0.4284	33.0192
0.0027	15.24	1250	0.4349	33.1229
0.0027	15.85	1300	0.4253	32.7598
0.0022	16.46	1350	0.4419	33.1143
0.0023	17.07	1400	0.4453	32.9154
0.002	17.68	1450	0.4457	32.5696
0.0014	18.29	1500	0.4592	32.8809
0.0014	18.9	1550	0.4757	32.8290
0.001	19.51	1600	0.4767	33.4169
0.0008	20.12	1650	0.4876	32.4831
0.0008	20.73	1700	0.4905	32.9760
0.0011	21.34	1750	0.4876	32.7252
0.0007	21.95	1800	0.4992	33.0105
0.0003	22.56	1850	0.5190	32.3102
0.0007	23.17	1900	0.5240	32.6734
0.0005	23.78	1950	0.5315	32.8809
0.0003	24.39	2000	0.5333	32.7771
0.0002	25.0	2050	0.5441	32.1200
0.0001	25.61	2100	0.5626	32.4313
0.0001	26.22	2150	0.5690	32.1546
0.0001	26.83	2200	0.5861	32.1978
0.0001	27.44	2250	0.6071	32.0163
0.0	28.05	2300	0.6214	32.6388
0.0001	28.66	2350	0.6333	32.7512
0.0	29.27	2400	0.6525	32.5782
0.0	29.88	2450	0.6627	32.6647
0.0	30.49	2500	0.6759	32.5523
0.0	31.1	2550	0.6960	33.3737
0.0	31.71	2600	0.7087	34.1864
0.0	32.32	2650	0.7228	34.4544
0.0	32.93	2700	0.7274	35.1634
0.0	33.54	2750	0.7327	35.7254
0.0	34.15	2800	0.7369	37.0569
0.0	34.76	2850	0.7405	38.2155
0.0	35.37	2900	0.7433	40.8871
0.0	35.98	2950	0.7441	41.6739
0.0	36.59	3000	0.7442	41.8554

Framework versions

Transformers 4.37.0.dev0
Pytorch 1.12.1
Datasets 2.16.1
Tokenizers 0.15.0

Chenxi-Chelsea-Liu/whisper-small-noisy-hindi-10dB

whisper-small-noisy-hindi-10dB

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

为此模型做种