Create README.md
Browse filesInitial draft of readme.md for hungarian
README.md
ADDED
@@ -0,0 +1,107 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- hu
|
4 |
+
license: apache-2.0
|
5 |
+
tags:
|
6 |
+
- automatic-speech-recognition
|
7 |
+
- mozilla-foundation/common_voice_7_0
|
8 |
+
- generated_from_trainer
|
9 |
+
- hu
|
10 |
+
- robust-speech-event
|
11 |
+
- model_for_talk
|
12 |
+
datasets:
|
13 |
+
- mozilla-foundation/common_voice_7_0
|
14 |
+
|
15 |
+
model-index:
|
16 |
+
- name: Akashpb13/xlsr_hungarian_new
|
17 |
+
results:
|
18 |
+
- task:
|
19 |
+
name: Automatic Speech Recognition
|
20 |
+
type: automatic-speech-recognition
|
21 |
+
dataset:
|
22 |
+
name: Common Voice 7
|
23 |
+
type: mozilla-foundation/common_voice_7_0
|
24 |
+
args: hu
|
25 |
+
metrics:
|
26 |
+
- name: Test WER
|
27 |
+
type: wer
|
28 |
+
value: 0.02698525418772714
|
29 |
+
- name: Test CER
|
30 |
+
type: cer
|
31 |
+
value: 0.005033063261641211
|
32 |
+
- task:
|
33 |
+
name: Automatic Speech Recognition
|
34 |
+
type: automatic-speech-recognition
|
35 |
+
dataset:
|
36 |
+
name: Robust Speech Event - Dev Data
|
37 |
+
type: speech-recognition-community-v2/dev_data
|
38 |
+
args: hu
|
39 |
+
metrics:
|
40 |
+
- name: Test WER
|
41 |
+
type: wer
|
42 |
+
value: 0.02698525418772714
|
43 |
+
- name: Test CER
|
44 |
+
type: cer
|
45 |
+
value: 0.005033063261641211
|
46 |
+
---
|
47 |
+
|
48 |
+
# Akashpb13/xlsr_hungarian_new
|
49 |
+
|
50 |
+
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - hu dataset.
|
51 |
+
It achieves the following results on evaluation set (which is 10 percent of train data set merged with invalidated data, reported, other, dev and validated datasets):
|
52 |
+
- Loss: 0.184265
|
53 |
+
- Wer: 0.292771
|
54 |
+
## Model description
|
55 |
+
More information needed
|
56 |
+
## Intended uses & limitations
|
57 |
+
More information needed
|
58 |
+
## Training and evaluation data
|
59 |
+
More information needed
|
60 |
+
## Training procedure
|
61 |
+
### Training hyperparameters
|
62 |
+
|
63 |
+
The following hyperparameters were used during training:
|
64 |
+
|
65 |
+
learning_rate: 0.000095637994662983496
|
66 |
+
train_batch_size: 16
|
67 |
+
eval_batch_size: 16
|
68 |
+
seed: 13
|
69 |
+
gradient_accumulation_steps: 16
|
70 |
+
total_train_batch_size: 316
|
71 |
+
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
72 |
+
lr_scheduler_type: cosine_with_restarts
|
73 |
+
lr_scheduler_warmup_steps: 500
|
74 |
+
num_epochs: 100
|
75 |
+
mixed_precision_training: Native AMP
|
76 |
+
|
77 |
+
|
78 |
+
### Training results
|
79 |
+
|
80 |
+
|
81 |
+
Step Training Loss Validation Loss Wer
|
82 |
+
500 4.825900 1.001413 0.810308
|
83 |
+
1000 0.561400 0.202275 0.361987
|
84 |
+
1500 0.298900 0.169643 0.326449
|
85 |
+
2000 0.236500 0.168602 0.316215
|
86 |
+
2500 0.199100 0.182484 0.308587
|
87 |
+
3000 0.179100 0.178076 0.303005
|
88 |
+
3500 0.161500 0.179107 0.299935
|
89 |
+
4000 0.151700 0.183371 0.295283
|
90 |
+
4500 0.143700 0.184443 0.295283
|
91 |
+
5000 0.138900 0.184265 0.292771
|
92 |
+
|
93 |
+
### Framework versions
|
94 |
+
- Transformers 4.16.0.dev0
|
95 |
+
- Pytorch 1.10.0+cu102
|
96 |
+
- Datasets 1.17.1.dev0
|
97 |
+
- Tokenizers 0.10.3
|
98 |
+
|
99 |
+
#### Evaluation Commands
|
100 |
+
|
101 |
+
1. To evaluate on `mozilla-foundation/common_voice_7_0` with split `test`
|
102 |
+
|
103 |
+
```bash
|
104 |
+
python eval.py --model_id Akashpb13/xlsr_hungarian_new --dataset mozilla-foundation/common_voice_7_0 --config hu --split test
|
105 |
+
```
|
106 |
+
|
107 |
+
|