Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,13 +1,10 @@
|
|
1 |
import gradio as gr
|
2 |
import librosa
|
3 |
import numpy as np
|
4 |
-
|
5 |
-
import
|
6 |
|
7 |
|
8 |
-
model_id = "AescF/hubert-base-ls960-finetuned-common_language"
|
9 |
-
processor = Wav2Vec2Processor.from_pretrained(model_id)
|
10 |
-
model = Wav2Vec2ForClassification.from_pretrained(model_id)
|
11 |
language_classes = {
|
12 |
0: "Arabic",
|
13 |
1: "Basque",
|
@@ -57,31 +54,51 @@ language_classes = {
|
|
57 |
}
|
58 |
|
59 |
|
60 |
-
def predict_language(audio):
|
61 |
-
# Read audio file
|
62 |
-
audio_input, sr = librosa.load(audio, sr=16000)
|
63 |
-
|
64 |
-
# Convert to suitable format
|
65 |
-
input_values = processor(audio_input, return_tensors="pt", padding=True).input_values
|
66 |
-
|
67 |
-
# Make prediction
|
68 |
-
with torch.no_grad():
|
69 |
-
logits = model(input_values).logits
|
70 |
-
|
71 |
-
# Compute probabilities
|
72 |
-
probabilities = torch.softmax(logits, dim=1)
|
73 |
-
|
74 |
-
# Retrieve label
|
75 |
-
predicted_language_idx = torch.argmax(probabilities[0]).item()
|
76 |
-
|
77 |
-
return {language_classes[predicted_language_idx]: float(probabilities[0][predicted_language_idx])}
|
78 |
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
85 |
)
|
86 |
-
|
87 |
-
iface.launch()
|
|
|
1 |
import gradio as gr
|
2 |
import librosa
|
3 |
import numpy as np
|
4 |
+
import torch
|
5 |
+
from transformers import pipeline
|
6 |
|
7 |
|
|
|
|
|
|
|
8 |
language_classes = {
|
9 |
0: "Arabic",
|
10 |
1: "Basque",
|
|
|
54 |
}
|
55 |
|
56 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
|
58 |
+
|
59 |
+
username = "AescF" ## Complete your username
|
60 |
+
model_id = "AescF/hubert-base-ls960-finetuned-common_language"
|
61 |
+
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
62 |
+
pipe = pipeline("audio-classification", model=model_id, device=device)
|
63 |
+
|
64 |
+
# def predict_trunc(filepath):
|
65 |
+
# preprocessed = pipe.preprocess(filepath)
|
66 |
+
# truncated = pipe.feature_extractor.pad(preprocessed,truncation=True, max_length = 16_000*30)
|
67 |
+
# model_outputs = pipe.forward(truncated)
|
68 |
+
# outputs = pipe.postprocess(model_outputs)
|
69 |
+
|
70 |
+
# return outputs
|
71 |
+
|
72 |
+
|
73 |
+
def classify_audio(filepath):
|
74 |
+
"""
|
75 |
+
Goes from
|
76 |
+
[{'score': 0.8339303731918335, 'label': 'country'},
|
77 |
+
{'score': 0.11914275586605072, 'label': 'rock'},]
|
78 |
+
to
|
79 |
+
{"country": 0.8339303731918335, "rock":0.11914275586605072}
|
80 |
+
"""
|
81 |
+
start_time = timer()
|
82 |
+
preds = pipe(filepath)
|
83 |
+
# preds = predict_trunc(filepath)
|
84 |
+
outputs = {}
|
85 |
+
pred_time = round(timer() - start_time, 5)
|
86 |
+
for p in preds:
|
87 |
+
outputs[p["label"]] = p["score"], timer
|
88 |
+
return outputs
|
89 |
+
|
90 |
+
|
91 |
+
title = "🎵 Music Genre Classifier"
|
92 |
+
description = """
|
93 |
+
Demo for a music genre classifier trained on [GTZAN](https://huggingface.co/datasets/marsyas/gtzan)
|
94 |
+
For more info checkout [GITHUB](https://github.com/AEscF)
|
95 |
+
"""
|
96 |
+
demo = gr.Interface(
|
97 |
+
fn=classify_audio,
|
98 |
+
inputs=gr.Audio(type="filepath"),
|
99 |
+
outputs=[gr.Label(label="Predictions"), gr.Number(label="Prediction time (s)")],
|
100 |
+
title=title,
|
101 |
+
description=description,
|
102 |
+
examples=filenames,
|
103 |
)
|
104 |
+
demo.launch()
|
|