Update README.md
Browse files
README.md
CHANGED
@@ -9,7 +9,77 @@ tags:
|
|
9 |
pipeline_tag: audio-classification
|
10 |
---
|
11 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
model_name_or_path = "quaja/hubert-base-amharic-speech-emotion-recognition"
|
13 |
config = AutoConfig.from_pretrained(model_name_or_path)
|
14 |
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(model_name_or_path)
|
15 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
pipeline_tag: audio-classification
|
10 |
---
|
11 |
|
12 |
+
import torch
|
13 |
+
import torch.nn as nn
|
14 |
+
import torch.nn.functional as F
|
15 |
+
import torchaudio
|
16 |
+
from transformers import AutoConfig, Wav2Vec2Processor
|
17 |
+
import librosa
|
18 |
+
import IPython.display as ipd
|
19 |
+
import numpy as np
|
20 |
+
import pandas as pd
|
21 |
+
|
22 |
model_name_or_path = "quaja/hubert-base-amharic-speech-emotion-recognition"
|
23 |
config = AutoConfig.from_pretrained(model_name_or_path)
|
24 |
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(model_name_or_path)
|
25 |
+
sampling_rate = feature_extractor.sampling_rate
|
26 |
+
model = HubertForSpeechClassification.from_pretrained(model_name_or_path)
|
27 |
+
|
28 |
+
def speech_file_to_array_fn(path, sampling_rate):
|
29 |
+
speech_array, _sampling_rate = torchaudio.load(path)
|
30 |
+
resampler = torchaudio.transforms.Resample(_sampling_rate)
|
31 |
+
speech = resampler(speech_array).squeeze().numpy()
|
32 |
+
return speech
|
33 |
+
|
34 |
+
|
35 |
+
def predict(path, sampling_rate):
|
36 |
+
speech = speech_file_to_array_fn(path, sampling_rate)
|
37 |
+
features = feature_extractor(speech, sampling_rate=sampling_rate, return_tensors="pt", padding=True)
|
38 |
+
|
39 |
+
input_values = features.input_values.to(device)
|
40 |
+
|
41 |
+
with torch.no_grad():
|
42 |
+
logits = model(input_values).logits
|
43 |
+
|
44 |
+
scores = F.softmax(logits, dim=1).detach().cpu().numpy()[0]
|
45 |
+
outputs = [{"Label": config.id2label[i], "Score": f"{round(score * 100, 3):.1f}%"} for i, score in enumerate(scores)]
|
46 |
+
return outputs
|
47 |
+
|
48 |
+
|
49 |
+
STYLES = """
|
50 |
+
<style>
|
51 |
+
div.display_data {
|
52 |
+
margin: 0 auto;
|
53 |
+
max-width: 500px;
|
54 |
+
}
|
55 |
+
table.xxx {
|
56 |
+
margin: 50px !important;
|
57 |
+
float: right !important;
|
58 |
+
clear: both !important;
|
59 |
+
}
|
60 |
+
table.xxx td {
|
61 |
+
min-width: 300px !important;
|
62 |
+
text-align: center !important;
|
63 |
+
}
|
64 |
+
</style>
|
65 |
+
""".strip()
|
66 |
+
|
67 |
+
def prediction(df_row):
|
68 |
+
path, label = df_row["path"], df_row["emotion"]
|
69 |
+
df = pd.DataFrame([{"Emotion": label, "Sentence": " "}])
|
70 |
+
setup = {
|
71 |
+
'border': 2,
|
72 |
+
'show_dimensions': True,
|
73 |
+
'justify': 'center',
|
74 |
+
'classes': 'xxx',
|
75 |
+
'escape': False,
|
76 |
+
}
|
77 |
+
ipd.display(ipd.HTML(STYLES + df.to_html(**setup) + "<br />"))
|
78 |
+
speech, sr = torchaudio.load(path)
|
79 |
+
resampler = torchaudio.transforms.Resample(sr)
|
80 |
+
speech = resampler(speech[0]).squeeze().numpy()
|
81 |
+
ipd.display(ipd.Audio(data=np.asarray(speech), autoplay=True, rate=sampling_rate))
|
82 |
+
|
83 |
+
outputs = predict(path, sampling_rate)
|
84 |
+
r = pd.DataFrame(outputs)
|
85 |
+
ipd.display(ipd.HTML(STYLES + r.to_html(**setup) + "<br />"))
|