saattrupdan
commited on
Commit
•
998bc0d
1
Parent(s):
4fd7a58
Update README.md
Browse files
README.md
CHANGED
@@ -1,105 +1,47 @@
|
|
1 |
---
|
|
|
|
|
2 |
license: apache-2.0
|
3 |
-
|
4 |
-
-
|
5 |
datasets:
|
6 |
-
-
|
|
|
|
|
7 |
model-index:
|
8 |
- name: wav2vec2-xls-r-300m-cv8-da
|
9 |
-
results:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
---
|
11 |
|
12 |
-
|
13 |
-
should probably proofread and complete it, then remove this comment. -->
|
14 |
-
|
15 |
-
# wav2vec2-xls-r-300m-cv8-da
|
16 |
-
|
17 |
-
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the common_voice dataset.
|
18 |
-
It achieves the following results on the evaluation set:
|
19 |
-
- Loss: 287.3684
|
20 |
-
- Wer: 0.2978
|
21 |
|
22 |
## Model description
|
23 |
|
24 |
-
|
25 |
-
|
26 |
-
## Intended uses & limitations
|
27 |
-
|
28 |
-
More information needed
|
29 |
-
|
30 |
-
## Training and evaluation data
|
31 |
-
|
32 |
-
More information needed
|
33 |
-
|
34 |
-
## Training procedure
|
35 |
-
|
36 |
-
### Training hyperparameters
|
37 |
-
|
38 |
-
The following hyperparameters were used during training:
|
39 |
-
- learning_rate: 4e-05
|
40 |
-
- train_batch_size: 4
|
41 |
-
- eval_batch_size: 8
|
42 |
-
- seed: 4242
|
43 |
-
- gradient_accumulation_steps: 8
|
44 |
-
- total_train_batch_size: 32
|
45 |
-
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
46 |
-
- lr_scheduler_type: linear
|
47 |
-
- lr_scheduler_warmup_steps: 500
|
48 |
-
- num_epochs: 500
|
49 |
-
- mixed_precision_training: Native AMP
|
50 |
-
|
51 |
-
### Training results
|
52 |
|
53 |
-
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
54 |
-
|:-------------:|:------:|:-----:|:---------------:|:------:|
|
55 |
-
| 736.2597 | 5.55 | 300 | 1418.4758 | 1.0 |
|
56 |
-
| 577.3605 | 11.11 | 600 | 1267.1588 | 1.0 |
|
57 |
-
| 476.434 | 16.66 | 900 | 842.5760 | 1.0028 |
|
58 |
-
| 215.5165 | 22.22 | 1200 | 356.9064 | 0.6405 |
|
59 |
-
| 138.4598 | 27.77 | 1500 | 256.0349 | 0.4889 |
|
60 |
-
| 104.3224 | 33.33 | 1800 | 228.9147 | 0.4435 |
|
61 |
-
| 82.2922 | 38.88 | 2100 | 208.3896 | 0.4134 |
|
62 |
-
| 68.0471 | 44.44 | 2400 | 205.3142 | 0.3976 |
|
63 |
-
| 57.6613 | 49.99 | 2700 | 208.4677 | 0.3935 |
|
64 |
-
| 52.6747 | 55.55 | 3000 | 213.9339 | 0.3928 |
|
65 |
-
| 43.4302 | 61.11 | 3300 | 212.9035 | 0.3808 |
|
66 |
-
| 40.6623 | 66.66 | 3600 | 221.2281 | 0.3850 |
|
67 |
-
| 34.9752 | 72.22 | 3900 | 224.3349 | 0.3795 |
|
68 |
-
| 33.6799 | 77.77 | 4200 | 218.9585 | 0.3696 |
|
69 |
-
| 31.5797 | 83.33 | 4500 | 211.9424 | 0.3713 |
|
70 |
-
| 28.2278 | 88.88 | 4800 | 226.6527 | 0.3705 |
|
71 |
-
| 26.6871 | 94.44 | 5100 | 227.9236 | 0.3689 |
|
72 |
-
| 25.8799 | 99.99 | 5400 | 227.0796 | 0.3653 |
|
73 |
-
| 25.0987 | 105.55 | 5700 | 231.7458 | 0.3640 |
|
74 |
-
| 22.328 | 111.11 | 6000 | 236.6981 | 0.3644 |
|
75 |
-
| 21.8799 | 116.66 | 6300 | 219.9040 | 0.3573 |
|
76 |
-
| 20.1069 | 122.22 | 6600 | 230.2603 | 0.3592 |
|
77 |
-
| 20.8641 | 127.77 | 6900 | 248.2307 | 0.3638 |
|
78 |
-
| 18.6306 | 133.33 | 7200 | 239.0962 | 0.3583 |
|
79 |
-
| 18.2617 | 138.88 | 7500 | 245.4144 | 0.3549 |
|
80 |
-
| 17.5175 | 144.44 | 7800 | 258.2551 | 0.3617 |
|
81 |
-
| 56.1866 | 149.99 | 8100 | 278.0422 | 0.3384 |
|
82 |
-
| 28.7438 | 155.55 | 8400 | 271.7619 | 0.3107 |
|
83 |
-
| 23.462 | 161.11 | 8700 | 279.4823 | 0.3091 |
|
84 |
-
| 20.5816 | 166.66 | 9000 | 274.7589 | 0.3070 |
|
85 |
-
| 18.7921 | 172.22 | 9300 | 265.1793 | 0.3024 |
|
86 |
-
| 16.5106 | 177.77 | 9600 | 269.0261 | 0.2977 |
|
87 |
-
| 17.1239 | 183.33 | 9900 | 261.8192 | 0.2956 |
|
88 |
-
| 15.4832 | 188.88 | 10200 | 271.2451 | 0.2982 |
|
89 |
-
| 16.2146 | 194.44 | 10500 | 269.5397 | 0.2957 |
|
90 |
-
| 13.7332 | 199.99 | 10800 | 276.0069 | 0.3017 |
|
91 |
-
| 13.914 | 205.55 | 11100 | 278.2491 | 0.3023 |
|
92 |
-
| 13.7551 | 211.11 | 11400 | 260.8341 | 0.2900 |
|
93 |
-
| 11.6575 | 216.66 | 11700 | 276.8361 | 0.2906 |
|
94 |
-
| 12.4438 | 222.22 | 12000 | 280.0102 | 0.2954 |
|
95 |
-
| 13.1577 | 227.77 | 12300 | 268.7986 | 0.2949 |
|
96 |
-
| 12.2083 | 233.33 | 12600 | 272.5989 | 0.2949 |
|
97 |
-
| 11.1147 | 238.88 | 12900 | 287.3684 | 0.2978 |
|
98 |
|
|
|
99 |
|
100 |
-
|
101 |
|
102 |
-
-
|
103 |
-
|
104 |
-
-
|
105 |
-
|
|
|
1 |
---
|
2 |
+
language:
|
3 |
+
- da
|
4 |
license: apache-2.0
|
5 |
+
tasks:
|
6 |
+
- automatic-speech-recognition
|
7 |
datasets:
|
8 |
+
- common_voice_8_0
|
9 |
+
metrics:
|
10 |
+
- wer
|
11 |
model-index:
|
12 |
- name: wav2vec2-xls-r-300m-cv8-da
|
13 |
+
results:
|
14 |
+
- task:
|
15 |
+
type: automatic-speech-recognition
|
16 |
+
dataset:
|
17 |
+
type: Alvenir/alvenir_asr_da_eval
|
18 |
+
name: Alvenir ASR test dataset
|
19 |
+
metrics:
|
20 |
+
- type: wer
|
21 |
+
value: 25.80
|
22 |
+
- task:
|
23 |
+
type: automatic-speech-recognition
|
24 |
+
dataset:
|
25 |
+
type: mozilla-foundation/common_voice_8_0
|
26 |
+
args: da
|
27 |
+
name: Danish Common Voice 8.0
|
28 |
+
metrics:
|
29 |
+
- type: wer
|
30 |
+
value: 26.45
|
31 |
---
|
32 |
|
33 |
+
# XLS-R-300m-CV8-da
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
|
35 |
## Model description
|
36 |
|
37 |
+
This model is a fine-tuned version of the multilingual acoustic model [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the Danish part of [Common Voice 8.0](https://huggingface.co/datasets/mozilla-foundation/common_voice_8_0), containing ~6 crowdsourced hours of read-aloud Danish speech.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
38 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
|
40 |
+
## Performance
|
41 |
|
42 |
+
The model achieves the following WER scores (lower is better):
|
43 |
|
44 |
+
| **Dataset** | **WER without LM** | **WER with 5-gram LM** |
|
45 |
+
| :---: | ---: | ---: |
|
46 |
+
| [Danish part of Common Voice 8.0](https://huggingface.co/datasets/mozilla-foundation/common_voice_8_0/viewer/da/train) | 31.33 | 26.45 |
|
47 |
+
| [Alvenir test set](https://huggingface.co/datasets/Alvenir/alvenir_asr_da_eval) | 30.54 | 25.80 |
|