Update README.md
Browse files
README.md
CHANGED
@@ -1,93 +1,93 @@
|
|
1 |
-
---
|
2 |
-
license: cc-by-4.0
|
3 |
-
tags:
|
4 |
-
- audio
|
5 |
-
- automatic-speech-recognition
|
6 |
-
- hf-asr-leaderboard
|
7 |
-
language: et
|
8 |
-
model-index:
|
9 |
-
- name: xls-r-300m-et
|
10 |
-
results:
|
11 |
-
- task:
|
12 |
-
name: Automatic Speech Recognition
|
13 |
-
type: automatic-speech-recognition
|
14 |
-
dataset:
|
15 |
-
name: Common Voice
|
16 |
-
type: common_voice
|
17 |
-
args: et
|
18 |
-
metrics:
|
19 |
-
- name: Test WER
|
20 |
-
type: wer
|
21 |
-
value:
|
22 |
-
- name: Test CER
|
23 |
-
type: cer
|
24 |
-
value:
|
25 |
-
- task:
|
26 |
-
name: Automatic Speech Recognition
|
27 |
-
type: automatic-speech-recognition
|
28 |
-
dataset:
|
29 |
-
name: Common Voice 8
|
30 |
-
type: mozilla-foundation/common_voice_8_0
|
31 |
-
args: et
|
32 |
-
metrics:
|
33 |
-
- name: Test WER
|
34 |
-
type: wer
|
35 |
-
value:
|
36 |
-
- name: Test CER
|
37 |
-
type: cer
|
38 |
-
value:
|
39 |
-
---
|
40 |
-
|
41 |
-
|
42 |
-
# XLS-R-300m-ET
|
43 |
-
|
44 |
-
This is a XLS-R-300M model [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) finetuned on around 800 hours of diverse Estonian data.
|
45 |
-
|
46 |
-
## Model description
|
47 |
-
This is a general-purpose Estonian ASR model trained in the Lab of Language Technology at TalTech. It consists of only the CTC-based end-to-end model, no language model is currently provided.
|
48 |
-
|
49 |
-
## Intended uses & limitations
|
50 |
-
|
51 |
-
This model is intended for general-purpose speech recognition, such as broadcast conversations, interviews, talks, etc.
|
52 |
-
|
53 |
-
## How to use
|
54 |
-
|
55 |
-
|
56 |
-
TODO
|
57 |
-
|
58 |
-
#### Limitations and bias
|
59 |
-
|
60 |
-
Since this model was trained on mostly broadcast speech and texts from the web, it might have problems correctly decoding the following:
|
61 |
-
* Speech containing technical and other domain-specific terms
|
62 |
-
* Children's speech
|
63 |
-
* Non-native speech
|
64 |
-
* Speech recorded under very noisy conditions or with a microphone far from the speaker
|
65 |
-
* Very spontaneous and overlapping speech
|
66 |
-
|
67 |
-
## Training data
|
68 |
-
Acoustic training data:
|
69 |
-
|
70 |
-
| Type | Amount (h) |
|
71 |
-
|-----------------------|:------:|
|
72 |
-
| Broadcast speech | 591 |
|
73 |
-
| Spontaneous speech | 53 |
|
74 |
-
| Elderly speech corpus | 53 |
|
75 |
-
| Talks, lectures | 49 |
|
76 |
-
| Parliament speeches | 31 |
|
77 |
-
| *Total* | *761* |
|
78 |
-
|
79 |
-
|
80 |
-
## Training procedure
|
81 |
-
|
82 |
-
Finetuned using Fairseq.
|
83 |
-
|
84 |
-
## Evaluation results
|
85 |
-
|
86 |
-
### WER
|
87 |
-
|
88 |
-
|Dataset | WER |
|
89 |
-
|---|---|
|
90 |
-
| jutusaated.devset | 7.9 |
|
91 |
-
| jutusaated.testset | 6.1 |
|
92 |
-
| Common Voice 6.1 | 12.5 |
|
93 |
-
| Common Voice 8.0 | 13.4 |
|
|
|
1 |
+
---
|
2 |
+
license: cc-by-4.0
|
3 |
+
tags:
|
4 |
+
- audio
|
5 |
+
- automatic-speech-recognition
|
6 |
+
- hf-asr-leaderboard
|
7 |
+
language: et
|
8 |
+
model-index:
|
9 |
+
- name: xls-r-300m-et
|
10 |
+
results:
|
11 |
+
- task:
|
12 |
+
name: Automatic Speech Recognition
|
13 |
+
type: automatic-speech-recognition
|
14 |
+
dataset:
|
15 |
+
name: Common Voice
|
16 |
+
type: common_voice
|
17 |
+
args: et
|
18 |
+
metrics:
|
19 |
+
- name: Test WER
|
20 |
+
type: wer
|
21 |
+
value: 12.520395591222402
|
22 |
+
- name: Test CER
|
23 |
+
type: cer
|
24 |
+
value: 2.7091152438624897
|
25 |
+
- task:
|
26 |
+
name: Automatic Speech Recognition
|
27 |
+
type: automatic-speech-recognition
|
28 |
+
dataset:
|
29 |
+
name: Common Voice 8
|
30 |
+
type: mozilla-foundation/common_voice_8_0
|
31 |
+
args: et
|
32 |
+
metrics:
|
33 |
+
- name: Test WER
|
34 |
+
type: wer
|
35 |
+
value: 13.38447882323104
|
36 |
+
- name: Test CER
|
37 |
+
type: cer
|
38 |
+
value: 2.9816686199500255
|
39 |
+
---
|
40 |
+
|
41 |
+
|
42 |
+
# XLS-R-300m-ET
|
43 |
+
|
44 |
+
This is a XLS-R-300M model [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) finetuned on around 800 hours of diverse Estonian data.
|
45 |
+
|
46 |
+
## Model description
|
47 |
+
This is a general-purpose Estonian ASR model trained in the Lab of Language Technology at TalTech. It consists of only the CTC-based end-to-end model, no language model is currently provided.
|
48 |
+
|
49 |
+
## Intended uses & limitations
|
50 |
+
|
51 |
+
This model is intended for general-purpose speech recognition, such as broadcast conversations, interviews, talks, etc.
|
52 |
+
|
53 |
+
## How to use
|
54 |
+
|
55 |
+
|
56 |
+
TODO
|
57 |
+
|
58 |
+
#### Limitations and bias
|
59 |
+
|
60 |
+
Since this model was trained on mostly broadcast speech and texts from the web, it might have problems correctly decoding the following:
|
61 |
+
* Speech containing technical and other domain-specific terms
|
62 |
+
* Children's speech
|
63 |
+
* Non-native speech
|
64 |
+
* Speech recorded under very noisy conditions or with a microphone far from the speaker
|
65 |
+
* Very spontaneous and overlapping speech
|
66 |
+
|
67 |
+
## Training data
|
68 |
+
Acoustic training data:
|
69 |
+
|
70 |
+
| Type | Amount (h) |
|
71 |
+
|-----------------------|:------:|
|
72 |
+
| Broadcast speech | 591 |
|
73 |
+
| Spontaneous speech | 53 |
|
74 |
+
| Elderly speech corpus | 53 |
|
75 |
+
| Talks, lectures | 49 |
|
76 |
+
| Parliament speeches | 31 |
|
77 |
+
| *Total* | *761* |
|
78 |
+
|
79 |
+
|
80 |
+
## Training procedure
|
81 |
+
|
82 |
+
Finetuned using Fairseq.
|
83 |
+
|
84 |
+
## Evaluation results
|
85 |
+
|
86 |
+
### WER
|
87 |
+
|
88 |
+
|Dataset | WER |
|
89 |
+
|---|---|
|
90 |
+
| jutusaated.devset | 7.9 |
|
91 |
+
| jutusaated.testset | 6.1 |
|
92 |
+
| Common Voice 6.1 | 12.5 |
|
93 |
+
| Common Voice 8.0 | 13.4 |
|