patrickvonplaten
commited on
Commit
•
f84b430
1
Parent(s):
db24822
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language: en
|
3 |
+
datasets:
|
4 |
+
- librispeech_asr
|
5 |
+
tags:
|
6 |
+
- audio
|
7 |
+
- automatic-speech-recognition
|
8 |
+
license: apache-2.0
|
9 |
+
---
|
10 |
+
|
11 |
+
TODO: [To be filled]
|
12 |
+
|
13 |
+
|
14 |
+
## Evaluation on LibriSpeech Test
|
15 |
+
|
16 |
+
The following script shows how to evaluate this model on the [LibriSpeech](https://huggingface.co/datasets/librispeech_asr) *"clean"* and *"other"* dataset.
|
17 |
+
|
18 |
+
```python
|
19 |
+
from datasets import load_dataset
|
20 |
+
from transformers import Speech2TextTransformerForConditionalGeneration, Speech2TextTransformerTokenizer
|
21 |
+
import soundfile as sf
|
22 |
+
from jiwer import wer
|
23 |
+
|
24 |
+
librispeech_eval = load_dataset("librispeech_asr", "clean", split="test") # change to "other" for other test dataset
|
25 |
+
|
26 |
+
model = Speech2TextTransformerForConditionalGeneration.from_pretrained("valhalla/s2t_librispeech_medium").to("cuda")
|
27 |
+
tokenizer = Speech2TextTransformerTokenizer.from_pretrained("valhalla/s2t_librispeech_medium", do_upper_case=True)
|
28 |
+
|
29 |
+
def map_to_array(batch):
|
30 |
+
speech, _ = sf.read(batch["file"])
|
31 |
+
batch["speech"] = speech
|
32 |
+
return batch
|
33 |
+
|
34 |
+
librispeech_eval = librispeech_eval.map(map_to_array)
|
35 |
+
|
36 |
+
def map_to_pred(batch):
|
37 |
+
features = tokenizer(batch["speech"], sample_rate=16000, padding=True, return_tensors="pt")
|
38 |
+
input_features = features.input_features.to("cuda")
|
39 |
+
attention_mask = features.attention_mask.to("cuda")
|
40 |
+
|
41 |
+
gen_tokens = model.generate(input_ids=input_features, attention_mask=attention_mask)
|
42 |
+
batch["transcription"] = tokenizer.batch_decode(gen_tokens, skip_special_tokens=True)
|
43 |
+
return batch
|
44 |
+
|
45 |
+
result = librispeech_eval.map(map_to_pred, batched=True, batch_size=8, remove_columns=["speech"])
|
46 |
+
|
47 |
+
print("WER:", wer(result["text"], result["transcription"]))
|
48 |
+
```
|
49 |
+
|
50 |
+
*Result (WER)*:
|
51 |
+
|
52 |
+
| "clean" | "other" |
|
53 |
+
|---|---|
|
54 |
+
| 3.5 | 7.8 |
|