patrickvonplaten
commited on
Commit
•
94d99c4
1
Parent(s):
413c384
update model card README.md
Browse files
README.md
ADDED
@@ -0,0 +1,90 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
tags:
|
3 |
+
- automatic-speech-recognition
|
4 |
+
- timit_asr
|
5 |
+
- generated_from_trainer
|
6 |
+
datasets:
|
7 |
+
- timit_asr
|
8 |
+
model-index:
|
9 |
+
- name: sat-base
|
10 |
+
results: []
|
11 |
+
---
|
12 |
+
|
13 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
14 |
+
should probably proofread and complete it, then remove this comment. -->
|
15 |
+
|
16 |
+
# sat-base
|
17 |
+
|
18 |
+
This model is a fine-tuned version of [microsoft/unispeech-sat-base](https://huggingface.co/microsoft/unispeech-sat-base) on the TIMIT_ASR - NA dataset.
|
19 |
+
It achieves the following results on the evaluation set:
|
20 |
+
- Loss: 0.7014
|
21 |
+
- Wer: 0.5374
|
22 |
+
|
23 |
+
## Model description
|
24 |
+
|
25 |
+
More information needed
|
26 |
+
|
27 |
+
## Intended uses & limitations
|
28 |
+
|
29 |
+
More information needed
|
30 |
+
|
31 |
+
## Training and evaluation data
|
32 |
+
|
33 |
+
More information needed
|
34 |
+
|
35 |
+
## Training procedure
|
36 |
+
|
37 |
+
### Training hyperparameters
|
38 |
+
|
39 |
+
The following hyperparameters were used during training:
|
40 |
+
- learning_rate: 3e-05
|
41 |
+
- train_batch_size: 32
|
42 |
+
- eval_batch_size: 1
|
43 |
+
- seed: 42
|
44 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
45 |
+
- lr_scheduler_type: linear
|
46 |
+
- lr_scheduler_warmup_steps: 1000
|
47 |
+
- num_epochs: 20.0
|
48 |
+
- mixed_precision_training: Native AMP
|
49 |
+
|
50 |
+
### Training results
|
51 |
+
|
52 |
+
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
53 |
+
|:-------------:|:-----:|:----:|:---------------:|:------:|
|
54 |
+
| 6.9958 | 0.69 | 100 | 6.7171 | 1.0 |
|
55 |
+
| 3.0453 | 1.38 | 200 | 3.0374 | 1.0 |
|
56 |
+
| 2.9989 | 2.07 | 300 | 2.9807 | 1.0 |
|
57 |
+
| 2.969 | 2.76 | 400 | 2.9579 | 1.0 |
|
58 |
+
| 2.903 | 3.45 | 500 | 2.9072 | 1.0 |
|
59 |
+
| 2.8565 | 4.14 | 600 | 2.8804 | 1.0 |
|
60 |
+
| 2.8195 | 4.83 | 700 | 2.7916 | 1.0 |
|
61 |
+
| 2.3134 | 5.52 | 800 | 2.1456 | 1.0004 |
|
62 |
+
| 1.5475 | 6.21 | 900 | 1.4663 | 0.9549 |
|
63 |
+
| 1.1295 | 6.9 | 1000 | 1.1140 | 0.7227 |
|
64 |
+
| 1.0181 | 7.59 | 1100 | 0.9258 | 0.6497 |
|
65 |
+
| 1.0252 | 8.28 | 1200 | 0.8430 | 0.6255 |
|
66 |
+
| 0.835 | 8.97 | 1300 | 0.8063 | 0.6032 |
|
67 |
+
| 0.662 | 9.66 | 1400 | 0.7595 | 0.5931 |
|
68 |
+
| 0.5558 | 10.34 | 1500 | 0.7322 | 0.5819 |
|
69 |
+
| 0.7596 | 11.03 | 1600 | 0.7120 | 0.5708 |
|
70 |
+
| 0.6169 | 11.72 | 1700 | 0.7073 | 0.5606 |
|
71 |
+
| 0.4565 | 12.41 | 1800 | 0.7124 | 0.5586 |
|
72 |
+
| 0.4554 | 13.1 | 1900 | 0.6880 | 0.5501 |
|
73 |
+
| 0.6216 | 13.79 | 2000 | 0.6783 | 0.5494 |
|
74 |
+
| 0.5393 | 14.48 | 2100 | 0.7067 | 0.5499 |
|
75 |
+
| 0.4095 | 15.17 | 2200 | 0.7014 | 0.5438 |
|
76 |
+
| 0.3551 | 15.86 | 2300 | 0.7000 | 0.5426 |
|
77 |
+
| 0.5112 | 16.55 | 2400 | 0.6866 | 0.5426 |
|
78 |
+
| 0.5139 | 17.24 | 2500 | 0.7134 | 0.5446 |
|
79 |
+
| 0.3638 | 17.93 | 2600 | 0.7130 | 0.5434 |
|
80 |
+
| 0.3327 | 18.62 | 2700 | 0.6980 | 0.5377 |
|
81 |
+
| 0.4385 | 19.31 | 2800 | 0.7017 | 0.5390 |
|
82 |
+
| 0.4986 | 20.0 | 2900 | 0.7014 | 0.5374 |
|
83 |
+
|
84 |
+
|
85 |
+
### Framework versions
|
86 |
+
|
87 |
+
- Transformers 4.12.0.dev0
|
88 |
+
- Pytorch 1.8.1
|
89 |
+
- Datasets 1.14.1.dev0
|
90 |
+
- Tokenizers 0.10.3
|