Upload inference.py
Browse files- inference.py +39 -0
inference.py
ADDED
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torch.nn.functional as F
|
3 |
+
import torchaudio
|
4 |
+
from transformers import AutoConfig, Wav2Vec2Processor
|
5 |
+
|
6 |
+
from Wav2Vec2ForSpeechClassification import Wav2Vec2ForSpeechClassification
|
7 |
+
|
8 |
+
MY_MODEL = "padmalcom/wav2vec2-large-nonverbalvocalization-classification"
|
9 |
+
|
10 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
11 |
+
config = AutoConfig.from_pretrained(MY_MODEL)
|
12 |
+
processor = Wav2Vec2Processor.from_pretrained(MY_MODEL)
|
13 |
+
sampling_rate = processor.feature_extractor.sampling_rate
|
14 |
+
model = Wav2Vec2ForSpeechClassification.from_pretrained(MY_MODEL).to(device)
|
15 |
+
|
16 |
+
def speech_file_to_array_fn(path, sampling_rate):
|
17 |
+
speech_array, _sampling_rate = torchaudio.load(path)
|
18 |
+
resampler = torchaudio.transforms.Resample(_sampling_rate)
|
19 |
+
speech = resampler(speech_array).squeeze().numpy()
|
20 |
+
return speech
|
21 |
+
|
22 |
+
|
23 |
+
def predict(path, sampling_rate):
|
24 |
+
speech = speech_file_to_array_fn(path, sampling_rate)
|
25 |
+
features = processor(speech, sampling_rate=sampling_rate, return_tensors="pt", padding=True)
|
26 |
+
|
27 |
+
input_values = features.input_values.to(device)
|
28 |
+
attention_mask = features.attention_mask.to(device)
|
29 |
+
|
30 |
+
with torch.no_grad():
|
31 |
+
logits = model(input_values, attention_mask=attention_mask).logits
|
32 |
+
|
33 |
+
scores = F.softmax(logits, dim=1).detach().cpu().numpy()[0]
|
34 |
+
outputs = [{"Vocalization": config.id2label[i], "Score": f"{round(score * 100, 3):.1f}%"} for i, score in enumerate(scores)]
|
35 |
+
return outputs
|
36 |
+
|
37 |
+
res = predict("test.wav", 16000)
|
38 |
+
max = max(res, key=lambda x: x['Score'])
|
39 |
+
print("Expected lip popping:", max)
|