Commit
•
fbe9c2d
1
Parent(s):
c735f34
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,47 +1,73 @@
|
|
| 1 |
---
|
| 2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
tags:
|
| 4 |
- automatic-speech-recognition
|
| 5 |
- speech
|
| 6 |
- Tensorflow
|
| 7 |
- Keras
|
| 8 |
- RNN
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
---
|
| 10 |
|
| 11 |
-
|
|
|
|
| 12 |
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
|
| 15 |
-
## Intended uses & limitations
|
| 16 |
-
|
| 17 |
-
More information needed
|
| 18 |
-
|
| 19 |
-
## Training and evaluation data
|
| 20 |
-
|
| 21 |
-
More information needed
|
| 22 |
-
|
| 23 |
-
## Training procedure
|
| 24 |
-
|
| 25 |
-
### Training hyperparameters
|
| 26 |
-
|
| 27 |
-
The following hyperparameters were used during training:
|
| 28 |
-
|
| 29 |
-
| Hyperparameters | Value |
|
| 30 |
-
| :-- | :-- |
|
| 31 |
-
| name | Adam |
|
| 32 |
-
| weight_decay | None |
|
| 33 |
-
| clipnorm | None |
|
| 34 |
-
| global_clipnorm | None |
|
| 35 |
-
| clipvalue | None |
|
| 36 |
-
| use_ema | False |
|
| 37 |
-
| ema_momentum | 0.99 |
|
| 38 |
-
| ema_overwrite_frequency | None |
|
| 39 |
-
| jit_compile | True |
|
| 40 |
-
| is_legacy_optimizer | False |
|
| 41 |
-
| learning_rate | 9.999999747378752e-05 |
|
| 42 |
-
| beta_1 | 0.9 |
|
| 43 |
-
| beta_2 | 0.999 |
|
| 44 |
-
| epsilon | 1e-07 |
|
| 45 |
-
| amsgrad | False |
|
| 46 |
-
| training_precision | float32 |
|
| 47 |
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
datasets:
|
| 3 |
+
- mozilla-foundation/common_voice_17_0
|
| 4 |
+
language: cs
|
| 5 |
+
library_name: tensorflow
|
| 6 |
+
license: cc-by-nc-sa-4.0
|
| 7 |
+
metrics:
|
| 8 |
+
- wer
|
| 9 |
+
- val_loss
|
| 10 |
+
pipeline_tag: automatic-speech-recognition
|
| 11 |
tags:
|
| 12 |
- automatic-speech-recognition
|
| 13 |
- speech
|
| 14 |
- Tensorflow
|
| 15 |
- Keras
|
| 16 |
- RNN
|
| 17 |
+
model-index:
|
| 18 |
+
- name: KobraSpeech RNN Czech
|
| 19 |
+
results:
|
| 20 |
+
- task:
|
| 21 |
+
type: speech-to-text
|
| 22 |
+
dataset:
|
| 23 |
+
name: mozilla-foundation/common_voice_17_0
|
| 24 |
+
type: common_voice
|
| 25 |
+
split: test
|
| 26 |
+
metrics:
|
| 27 |
+
- type: wer
|
| 28 |
+
value: '0.6982'
|
| 29 |
---
|
| 30 |
|
| 31 |
+
# KobraSpeech RNN Czech
|
| 32 |
+
This is a lightweight speech-to-text model for Czech language. It was trained on the Common Voice dataset.
|
| 33 |
|
| 34 |
+
## Training progress
|
| 35 |
+
| Epoch | Loss | Val Loss |
|
| 36 |
+
| --- | --- | --- |
|
| 37 |
+
| 1 | 145.0826 | 101.9806 |
|
| 38 |
+
| 2 | 88.5889 | 80.9404 |
|
| 39 |
+
| 3 | 71.0080 | 72.7689 |
|
| 40 |
+
| 4 | 61.9973 | 68.7629 |
|
| 41 |
+
| 5 | 56.7657 | 60.8039 |
|
| 42 |
+
| 6 | 51.5836 | 56.6200 |
|
| 43 |
+
| 7 | 47.6242 | 58.4478 |
|
| 44 |
+
| 8 | 44.3805 | 54.3059 |
|
| 45 |
+
| 9 | 41.5582 | 49.7450 |
|
| 46 |
+
| 10 | 39.1244 | 51.0741 |
|
| 47 |
+
| 11 | 36.9500 | 46.6725 |
|
| 48 |
+
| 12 | 35.0127 | 45.6165 |
|
| 49 |
+
| 13 | 33.2974 | 47.7714 |
|
| 50 |
+
| 14 | 31.6605 | 45.0911 |
|
| 51 |
+
| 15 | 30.0918 | 43.3004 |
|
| 52 |
+
| 16 | 28.8173 | 42.9870 |
|
| 53 |
+
| 17 | 27.5169 | 42.2732 |
|
| 54 |
+
| 18 | 26.3582 | 42.9355 |
|
| 55 |
+
| 19 | 25.2368 | 42.0441 |
|
| 56 |
+
| 20 | 24.2527 | 41.2783 |
|
| 57 |
+
| 21 | 23.3302 | 40.5552 |
|
| 58 |
+
| 22 | 22.3662 | 42.3867 |
|
| 59 |
+
| 23 | 21.5657 | 41.0113 |
|
| 60 |
+
| 24 | 20.7213 | 42.3488 |
|
| 61 |
+
| 25 | 19.9843 | 41.7464 |
|
| 62 |
+
| 26 | 22.3809 | 40.7493 |
|
| 63 |
+
| 27 | 21.5943 | 40.4331 |
|
| 64 |
+
| 28 | 20.6919 | 41.5385 |
|
| 65 |
+
| 29 | 19.9768 | 41.5923 |
|
| 66 |
+
| 30 | 19.2961 | 39.0283 |
|
| 67 |
+
| 31 | 18.6037 | 40.4818 |
|
| 68 |
+
| 32 | 17.9178 | 40.1532 |
|
| 69 |
+
| 33 | 17.3384 | 40.9723 |
|
| 70 |
+
| 34 | 16.7528 | 39.4724 |
|
| 71 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 72 |
|
| 73 |
+
This model was created and trained by [Kobrasoft](https://kobrasoft.cz)
|