Commit ·
b1bd899
1
Parent(s): 32138f2
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,78 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- it
|
| 4 |
+
license: apache-2.0
|
| 5 |
+
tags:
|
| 6 |
+
- hf-asr-leaderboard
|
| 7 |
+
- generated_from_trainer
|
| 8 |
+
datasets:
|
| 9 |
+
- mozilla-foundation/common_voice_11_0
|
| 10 |
+
metrics:
|
| 11 |
+
- wer
|
| 12 |
+
model-index:
|
| 13 |
+
- name: Whisper Tiny it 5
|
| 14 |
+
results:
|
| 15 |
+
- task:
|
| 16 |
+
name: Automatic Speech Recognition
|
| 17 |
+
type: automatic-speech-recognition
|
| 18 |
+
dataset:
|
| 19 |
+
name: Common Voice 11.0
|
| 20 |
+
type: mozilla-foundation/common_voice_11_0
|
| 21 |
+
config: it
|
| 22 |
+
split: test[:10%]
|
| 23 |
+
args: 'config: it, split: test'
|
| 24 |
+
metrics:
|
| 25 |
+
- name: Wer
|
| 26 |
+
type: wer
|
| 27 |
+
value: 41.271491957848035
|
| 28 |
+
---
|
| 29 |
+
|
| 30 |
+
# Whisper Tiny it 5
|
| 31 |
+
|
| 32 |
+
This model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) on the Common Voice 11.0 dataset.
|
| 33 |
+
It achieves the following results on the evaluation set:
|
| 34 |
+
- Loss: 0.760934
|
| 35 |
+
- Wer: 41.271492
|
| 36 |
+
|
| 37 |
+
## Model description
|
| 38 |
+
|
| 39 |
+
This model is the openai whisper small transformer adapted for Italian audio to text transcription. This model has weight decay set to 0.1 and the learning rate has been set to 1e-4 in the hyperparameter tuning process and it improved the performance on the evaluation set.
|
| 40 |
+
|
| 41 |
+
## Intended uses & limitations
|
| 42 |
+
|
| 43 |
+
The model is available through its [HuggingFace web app](https://huggingface.co/spaces/GIanlucaRub/whisper-it)
|
| 44 |
+
|
| 45 |
+
## Training and evaluation data
|
| 46 |
+
|
| 47 |
+
Data used for training is the initial 10% of train and validation of [Italian Common Voice](https://huggingface.co/datasets/mozilla-foundation/common_voice_11_0/viewer/it/train) 11.0 from Mozilla Foundation.
|
| 48 |
+
The dataset used for evaluation is the initial 10% of test of Italian Common Voice.
|
| 49 |
+
|
| 50 |
+
|
| 51 |
+
## Training procedure
|
| 52 |
+
|
| 53 |
+
After loading the pre trained model, it has been trained on the dataset.
|
| 54 |
+
|
| 55 |
+
### Training hyperparameters
|
| 56 |
+
|
| 57 |
+
The following hyperparameters were used during training:
|
| 58 |
+
- learning_rate: 1e-04
|
| 59 |
+
- train_batch_size: 16
|
| 60 |
+
- eval_batch_size: 8
|
| 61 |
+
- seed: 42
|
| 62 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 63 |
+
- lr_scheduler_type: linear
|
| 64 |
+
- lr_scheduler_warmup_steps: 500
|
| 65 |
+
- training_steps: 4000
|
| 66 |
+
- mixed_precision_training: Native AMP
|
| 67 |
+
### Training results
|
| 68 |
+
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
| 69 |
+
|:-------------:|:-----:|:----:|:---------------:|:-------:|
|
| 70 |
+
| 0.7015 | 0.95 | 1000 | 0.9463 | 64.4689 |
|
| 71 |
+
| 0.3579 | 1.91 | 2000 | 0.8363 | 51.7471 |
|
| 72 |
+
| 0.1388 | 2.86 | 3000 | 0.7766 | 43.6425 |
|
| 73 |
+
| 0.0403 | 3.82 | 4000 | 0.7609 | 41.2715 |
|
| 74 |
+
### Framework versions
|
| 75 |
+
- Transformers 4.26.0.dev0
|
| 76 |
+
- Pytorch 1.12.1+cu113
|
| 77 |
+
- Datasets 2.7.1
|
| 78 |
+
- Tokenizers 0.13.2
|