Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- tt
|
| 4 |
+
license: apache-2.0
|
| 5 |
+
tags:
|
| 6 |
+
- whisper-event
|
| 7 |
+
datasets:
|
| 8 |
+
- mozilla-foundation/common_voice_11_0
|
| 9 |
+
metrics:
|
| 10 |
+
- wer
|
| 11 |
+
model-index:
|
| 12 |
+
- name: Whisper Tiny Tatar - Kirill Milintsevich
|
| 13 |
+
results:
|
| 14 |
+
- task:
|
| 15 |
+
name: Automatic Speech Recognition
|
| 16 |
+
type: automatic-speech-recognition
|
| 17 |
+
dataset:
|
| 18 |
+
name: mozilla-foundation/common_voice_11_0 tt
|
| 19 |
+
type: mozilla-foundation/common_voice_11_0
|
| 20 |
+
config: tt
|
| 21 |
+
split: test
|
| 22 |
+
args: tt
|
| 23 |
+
metrics:
|
| 24 |
+
- name: Wer
|
| 25 |
+
type: wer
|
| 26 |
+
value: 49.228482446206115
|
| 27 |
+
---
|
| 28 |
+
|
| 29 |
+
# Whisper Small Tatar - Kirill Milintsevich
|
| 30 |
+
|
| 31 |
+
This model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) on the Common Voice 11.0 dataset.
|
| 32 |
+
It achieves the following results on the evaluation set:
|
| 33 |
+
- Loss: 0.5106
|
| 34 |
+
- Wer: 49.2285
|
| 35 |
+
|
| 36 |
+
## Model description
|
| 37 |
+
|
| 38 |
+
More information needed
|
| 39 |
+
|
| 40 |
+
## Intended uses & limitations
|
| 41 |
+
|
| 42 |
+
More information needed
|
| 43 |
+
|
| 44 |
+
## Training and evaluation data
|
| 45 |
+
|
| 46 |
+
More information needed
|
| 47 |
+
|
| 48 |
+
## Training procedure
|
| 49 |
+
|
| 50 |
+
### Training hyperparameters
|
| 51 |
+
|
| 52 |
+
The following hyperparameters were used during training:
|
| 53 |
+
- learning_rate: 1e-05
|
| 54 |
+
- train_batch_size: 64
|
| 55 |
+
- eval_batch_size: 32
|
| 56 |
+
- seed: 42
|
| 57 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 58 |
+
- lr_scheduler_type: linear
|
| 59 |
+
- lr_scheduler_warmup_steps: 500
|
| 60 |
+
- training_steps: 5000
|
| 61 |
+
- mixed_precision_training: Native AMP
|
| 62 |
+
### Training results
|
| 63 |
+
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
| 64 |
+
|:-------------:|:-----:|:----:|:---------------:|:-------:|
|
| 65 |
+
| 0.4268 | 2.49 | 500 | 0.6232 | 63.6537 |
|
| 66 |
+
| 0.2331 | 4.98 | 1000 | 0.5044 | 52.3818 |
|
| 67 |
+
| 0.1332 | 7.46 | 1500 | 0.4927 | 50.2300 |
|
| 68 |
+
| 0.09 | 9.95 | 2000 | 0.5106 | 49.2285 |
|
| 69 |
+
| 0.048 | 12.44 | 2500 | 0.5526 | 49.7806 |
|
| 70 |
+
| 0.0346 | 14.93 | 3000 | 0.5850 | 50.0319 |
|
| 71 |
+
| 0.0181 | 17.41 | 3500 | 0.6276 | 50.5592 |
|
| 72 |
+
| 0.0122 | 19.9 | 4000 | 0.6494 | 50.3327 |
|
| 73 |
+
| 0.0086 | 22.39 | 4500 | 0.6737 | 50.6688 |
|
| 74 |
+
| 0.0077 | 24.88 | 5000 | 0.6777 | 50.6724 |
|