Update README.md
Browse files
README.md
CHANGED
@@ -1 +1,89 @@
|
|
1 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- sv-SE
|
4 |
+
license: apache-2.0
|
5 |
+
tags:
|
6 |
+
- automatic-speech-recognition
|
7 |
+
- mozilla-foundation/common_voice_7_0
|
8 |
+
- generated_from_trainer
|
9 |
+
- nl
|
10 |
+
- robust-speech-event
|
11 |
+
- model_for_talk
|
12 |
+
datasets:
|
13 |
+
- mozilla-foundation/common_voice_7_0
|
14 |
+
model-index:
|
15 |
+
- name: XLS-R-300M - Dutch
|
16 |
+
results:
|
17 |
+
- task:
|
18 |
+
name: Automatic Speech Recognition
|
19 |
+
type: automatic-speech-recognition
|
20 |
+
dataset:
|
21 |
+
name: Common Voice 7
|
22 |
+
type: mozilla-foundation/common_voice_7_0
|
23 |
+
args: nl
|
24 |
+
metrics:
|
25 |
+
- name: Test WER
|
26 |
+
type: wer
|
27 |
+
value: ???
|
28 |
+
- name: Test CER
|
29 |
+
type: cer
|
30 |
+
value: ???
|
31 |
+
- task:
|
32 |
+
name: Automatic Speech Recognition
|
33 |
+
type: automatic-speech-recognition
|
34 |
+
dataset:
|
35 |
+
name: Robust Speech Event - Dev Data
|
36 |
+
type: speech-recognition-community-v2/dev_data
|
37 |
+
args: sv
|
38 |
+
metrics:
|
39 |
+
- name: Test WER
|
40 |
+
type: wer
|
41 |
+
value: ???
|
42 |
+
- name: Test CER
|
43 |
+
type: cer
|
44 |
+
value: ???
|
45 |
+
---
|
46 |
+
|
47 |
+
# xlsr300m_cv_8.0_nl
|
48 |
+
|
49 |
+
|
50 |
+
#### Evaluation Commands
|
51 |
+
1. To evaluate on `mozilla-foundation/common_voice_7_0` with split `test`
|
52 |
+
|
53 |
+
```bash
|
54 |
+
python eval.py --model_id Iskaj/xlsr300m_cv_8.0_nl --dataset mozilla-foundation/common_voice_8_0 --config nl --split test
|
55 |
+
```
|
56 |
+
|
57 |
+
2. To evaluate on `speech-recognition-community-v2/dev_data`
|
58 |
+
|
59 |
+
```bash
|
60 |
+
python eval.py --model_id Iskaj/xlsr300m_cv_8.0_nl --dataset speech-recognition-community-v2/dev_data --config nl --split validation --chunk_length_s 5.0 --stride_length_s 1.0
|
61 |
+
```
|
62 |
+
|
63 |
+
### Inference
|
64 |
+
|
65 |
+
```python
|
66 |
+
import torch
|
67 |
+
from datasets import load_dataset
|
68 |
+
from transformers import AutoModelForCTC, AutoProcessor
|
69 |
+
import torchaudio.functional as F
|
70 |
+
|
71 |
+
model_id = "Iskaj/xlsr300m_cv_8.0_nl"
|
72 |
+
|
73 |
+
sample_iter = iter(load_dataset("mozilla-foundation/common_voice_8_0", "nl", split="test", streaming=True, use_auth_token=True))
|
74 |
+
|
75 |
+
sample = next(sample_iter)
|
76 |
+
resampled_audio = F.resample(torch.tensor(sample["audio"]["array"]), 48_000, 16_000).numpy()
|
77 |
+
|
78 |
+
model = AutoModelForCTC.from_pretrained(model_id)
|
79 |
+
processor = AutoProcessor.from_pretrained(model_id)
|
80 |
+
|
81 |
+
inputs = processor(resampled_audio, sampling_rate=16_000, return_tensors="pt")
|
82 |
+
with torch.no_grad():
|
83 |
+
logits = model(**inputs).logits
|
84 |
+
predicted_ids = torch.argmax(logits, dim=-1)
|
85 |
+
transcription = processor.batch_decode(predicted_ids)
|
86 |
+
|
87 |
+
transcription[0].lower()
|
88 |
+
#'het kontine schip lag aangemeert in de aven'
|
89 |
+
```
|