Update README.md
Browse files
README.md
CHANGED
@@ -3,18 +3,18 @@ license: apache-2.0
|
|
3 |
---
|
4 |
|
5 |
# wav2vec2-base-da-ft-nst
|
6 |
-
This
|
7 |
|
8 |
The model was trained using fairseq and then converted to huggingface/transformers format.
|
9 |
|
10 |
-
Alvenir is always happy to help with your own open-source ASR projects
|
11 |
|
12 |
## Usage
|
13 |
```Python
|
14 |
import soundfile as sf
|
15 |
import torch
|
16 |
|
17 |
-
from transformers import Wav2Vec2CTCTokenizer, Wav2Vec2Tokenizer,
|
18 |
Wav2Vec2ForCTC
|
19 |
|
20 |
|
@@ -22,10 +22,6 @@ def get_tokenizer(model_path: str) -> Wav2Vec2CTCTokenizer:
|
|
22 |
return Wav2Vec2Tokenizer.from_pretrained(model_path)
|
23 |
|
24 |
|
25 |
-
def get_feature_extractor(model_path: str) -> Wav2Vec2FeatureExtractor:
|
26 |
-
return Wav2Vec2FeatureExtractor.from_pretrained(model_path)
|
27 |
-
|
28 |
-
|
29 |
def get_processor(model_path: str) -> Wav2Vec2Processor:
|
30 |
return Wav2Vec2Processor.from_pretrained(model_path)
|
31 |
|
@@ -55,8 +51,10 @@ print(transcription)
|
|
55 |
|
56 |
```
|
57 |
## Benchmark results
|
58 |
-
|
|
|
|
|
59 |
|---------------------|------------|--------------------|
|
60 |
| NST test | 15,8% | 11.9% |
|
61 |
-
| alvenir-asr-da-eval |
|
62 |
-
|
|
|
|
3 |
---
|
4 |
|
5 |
# wav2vec2-base-da-ft-nst
|
6 |
+
This the [alvenir wav2vec2 model](https://huggingface.co/Alvenir/wav2vec2-base-da) for Danish ASR finetuned by Alvenir on the public NST dataset. The model is trained on 16kHz, so make sure your data is the same sample rate.
|
7 |
|
8 |
The model was trained using fairseq and then converted to huggingface/transformers format.
|
9 |
|
10 |
+
Alvenir is always happy to help with your own open-source ASR projects, customized domain specializations or premium models. ;-)
|
11 |
|
12 |
## Usage
|
13 |
```Python
|
14 |
import soundfile as sf
|
15 |
import torch
|
16 |
|
17 |
+
from transformers import Wav2Vec2CTCTokenizer, Wav2Vec2Tokenizer, Wav2Vec2Processor, \
|
18 |
Wav2Vec2ForCTC
|
19 |
|
20 |
|
|
|
22 |
return Wav2Vec2Tokenizer.from_pretrained(model_path)
|
23 |
|
24 |
|
|
|
|
|
|
|
|
|
25 |
def get_processor(model_path: str) -> Wav2Vec2Processor:
|
26 |
return Wav2Vec2Processor.from_pretrained(model_path)
|
27 |
|
|
|
51 |
|
52 |
```
|
53 |
## Benchmark results
|
54 |
+
This is some benchmark results on the public available datasets in Danish.
|
55 |
+
|
56 |
+
| Dataset | WER Greedy | WER with Language Model |
|
57 |
|---------------------|------------|--------------------|
|
58 |
| NST test | 15,8% | 11.9% |
|
59 |
+
| alvenir-asr-da-eval | 19.0% | 12.1% |
|
60 |
+
| common_voice_80 da test | 26,3% | ?? |
|