README.md

metadata

language:
  - bn
license: apache-2.0
tags:
  - whisper-event
  - generated_from_trainer
datasets:
  - mozilla-foundation/common_voice_11_0
  - google/fleurs
  - openslr
metrics:
  - wer
model-index:
  - name: Whisper Small - Mohammed Rakib
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: mozilla-foundation/common_voice_11_0
          type: mozilla-foundation/common_voice_11_0
          config: bn
          split: test
        metrics:
          - type: wer
            value: 10.09
            name: WER
          - type: cer
            value: 5.31
            name: CER
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: google/fleurs
          type: google/fleurs
          config: bn_in
          split: test
        metrics:
          - type: wer
            value: 14.74
            name: WER
          - type: cer
            value: 8.87
            name: CER

Whisper Small - Mohammed Rakib

This model is a fine-tuned version of openai/whisper-small on the common-voice-11, the google-fleurs and the openslr53 datasets. It achieves the following results on the evaluation set:

• Loss: 0.0605
• Cer: 5.6737
• Wer: 10.5971

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: 4
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 128
• total_train_batch_size: 512
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 888
• training_steps: 3000

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.4902	0.11	100	1.2636	1208.7378	889.9406
0.5703	0.23	200	0.4612	121.7766	108.6401
0.3679	0.34	300	0.3046	17.7385	35.2744
0.2301	0.45	400	0.2059	14.8853	29.7671
0.191	0.56	500	0.1693	12.7561	25.4528
0.1605	0.68	600	0.1462	11.5880	22.7845
0.146	0.79	700	0.1300	10.4321	20.5541
0.1296	0.9	800	0.1156	9.8572	19.2143
0.1212	1.01	900	0.1055	8.9462	17.4004
0.1072	1.13	1000	0.0978	8.2675	15.9234
0.1013	1.24	1100	0.0912	7.7918	15.0605
0.0952	1.35	1200	0.0854	7.5497	14.3207
0.0915	1.47	1300	0.0809	6.9833	13.3163
0.0843	1.58	1400	0.0780	6.6422	12.7179
0.0819	1.69	1500	0.0744	6.7287	12.6589
0.0798	1.8	1600	0.0718	6.4962	12.3022
0.0774	1.92	1700	0.0694	6.2198	11.8414
0.0695	2.03	1800	0.0680	6.1346	11.5683
0.0686	2.14	1900	0.0662	5.9758	11.2485
0.0681	2.25	2000	0.0647	6.0599	11.2842
0.0661	2.37	2100	0.0639	5.9500	11.1356
0.0653	2.48	2200	0.0631	5.8114	10.8952
0.0636	2.59	2300	0.0622	5.8502	10.9385
0.0641	2.71	2400	0.0615	5.7382	10.7015
0.0633	2.82	2500	0.0612	5.7038	10.6455
0.0626	2.93	2600	0.0608	5.8058	10.7597
0.06	3.04	2700	0.0605	5.7328	10.6374
0.0584	3.16	2800	0.0605	5.6737	10.5971
0.0585	3.27	2900	0.0604	5.6877	10.6098
0.0598	3.38	3000	0.0603	5.7075	10.6043

Framework versions

• Transformers 4.27.0.dev0
• Pytorch 1.13.1+cu117
• Datasets 2.10.1
• Tokenizers 0.13.2