Spaces:
Running
Running
Commit
·
0474f44
1
Parent(s):
f26cacc
Added files
Browse files
app.py
CHANGED
@@ -3,10 +3,9 @@ import torch
|
|
3 |
import torchaudio
|
4 |
import numpy as np
|
5 |
import plotly.graph_objs as go
|
6 |
-
import os #
|
7 |
-
from
|
8 |
-
from
|
9 |
-
from audio_dataset import pad_audio
|
10 |
|
11 |
app = Flask(__name__)
|
12 |
|
@@ -16,17 +15,7 @@ model = BoundaryDetectionModel().to(device)
|
|
16 |
model.load_state_dict(torch.load("checkpoint_epoch_21_eer_0.24.pth", map_location=device)["model_state_dict"])
|
17 |
model.eval()
|
18 |
|
19 |
-
def convert_to_wav(audio_path, temp_path="temp_audio.wav"):
|
20 |
-
# Check if the file is already in .wav format
|
21 |
-
if audio_path.lower().endswith(".wav"):
|
22 |
-
return audio_path
|
23 |
-
# Convert to .wav using pydub if it's not already in .wav
|
24 |
-
audio = AudioSegment.from_file(audio_path)
|
25 |
-
audio.export(temp_path, format="wav")
|
26 |
-
return temp_path
|
27 |
-
|
28 |
def preprocess_audio(audio_path, sample_rate=16000, target_length=8):
|
29 |
-
# Load the audio waveform
|
30 |
waveform, sr = torchaudio.load(audio_path)
|
31 |
if sr != sample_rate:
|
32 |
waveform = torchaudio.transforms.Resample(sr, sample_rate)(waveform)
|
@@ -52,10 +41,8 @@ def predict():
|
|
52 |
if file.filename == '':
|
53 |
return "No selected file", 400
|
54 |
|
55 |
-
#
|
56 |
-
|
57 |
-
file.save(original_path)
|
58 |
-
file_path = convert_to_wav(original_path) # Convert to .wav if needed
|
59 |
|
60 |
# Preprocess audio and perform inference
|
61 |
audio_tensor = preprocess_audio(file_path)
|
@@ -96,8 +83,9 @@ def predict():
|
|
96 |
def return_to_index():
|
97 |
# Delete temporary files before returning to index
|
98 |
try:
|
99 |
-
os.remove("
|
100 |
-
|
|
|
101 |
except OSError as e:
|
102 |
print(f"Error deleting temporary files: {e}")
|
103 |
|
@@ -169,4 +157,4 @@ def plot_fake_frames_waveform(output, prediction_flat, waveform, fake_frame_inte
|
|
169 |
return plot_html
|
170 |
|
171 |
if __name__ == '__main__':
|
172 |
-
app.run()
|
|
|
3 |
import torchaudio
|
4 |
import numpy as np
|
5 |
import plotly.graph_objs as go
|
6 |
+
import os # Import os for file operations
|
7 |
+
from model import BoundaryDetectionModel # Assuming your model is defined here
|
8 |
+
from audio_dataset import pad_audio # Assuming you have a function to pad audio
|
|
|
9 |
|
10 |
app = Flask(__name__)
|
11 |
|
|
|
15 |
model.load_state_dict(torch.load("checkpoint_epoch_21_eer_0.24.pth", map_location=device)["model_state_dict"])
|
16 |
model.eval()
|
17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
def preprocess_audio(audio_path, sample_rate=16000, target_length=8):
|
|
|
19 |
waveform, sr = torchaudio.load(audio_path)
|
20 |
if sr != sample_rate:
|
21 |
waveform = torchaudio.transforms.Resample(sr, sample_rate)(waveform)
|
|
|
41 |
if file.filename == '':
|
42 |
return "No selected file", 400
|
43 |
|
44 |
+
file_path = "temp_audio.wav" # Temporary file to store uploaded audio
|
45 |
+
file.save(file_path)
|
|
|
|
|
46 |
|
47 |
# Preprocess audio and perform inference
|
48 |
audio_tensor = preprocess_audio(file_path)
|
|
|
83 |
def return_to_index():
|
84 |
# Delete temporary files before returning to index
|
85 |
try:
|
86 |
+
os.remove("temp_audio.wav") # Remove the temporary audio file
|
87 |
+
# If you have any other temporary files (like plots), remove them here too.
|
88 |
+
# Example: os.remove("temp_plot.html") if you save plots as HTML files.
|
89 |
except OSError as e:
|
90 |
print(f"Error deleting temporary files: {e}")
|
91 |
|
|
|
157 |
return plot_html
|
158 |
|
159 |
if __name__ == '__main__':
|
160 |
+
app.run()
|