Shokoufehhh
commited on
Update app.py
Browse files
app.py
CHANGED
@@ -1,7 +1,72 @@
|
|
|
|
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torchaudio
|
3 |
import gradio as gr
|
4 |
+
from sgmse.model import ScoreModel
|
5 |
+
from sgmse.util.other import pad_spec
|
6 |
+
import time # Import the time module
|
7 |
+
import os
|
8 |
|
9 |
+
# Define parameters based on the configuration in enhancement.py
|
10 |
+
args = {
|
11 |
+
"test_dir": "./test_data", # example directory, adjust as needed
|
12 |
+
"enhanced_dir": "./enhanced_data", # example directory, adjust as needed
|
13 |
+
"ckpt": "https://huggingface.co/sp-uhh/speech-enhancement-sgmse/resolve/main/train_vb_29nqe0uh_epoch%3D115.ckpt",
|
14 |
+
"corrector": "ald",
|
15 |
+
"corrector_steps": 1,
|
16 |
+
"snr": 0.5,
|
17 |
+
"N": 30,
|
18 |
+
"device": "cuda" if torch.cuda.is_available() else "cpu"
|
19 |
+
}
|
20 |
|
21 |
+
# Ensure the model is loaded to the correct device
|
22 |
+
model = ScoreModel.load_from_checkpoint(args["ckpt"]).to(args["device"])
|
23 |
+
|
24 |
+
def enhance_speech(audio_file):
|
25 |
+
start_time = time.time() # Start the timer
|
26 |
+
|
27 |
+
# Load and process the audio file
|
28 |
+
y, sr = torchaudio.load(audio_file) # Gradio passes the file path
|
29 |
+
print(f"Loaded audio in {time.time() - start_time:.2f}s")
|
30 |
+
T_orig = y.size(1)
|
31 |
+
|
32 |
+
# Normalize
|
33 |
+
norm_factor = y.abs().max()
|
34 |
+
y = y / norm_factor
|
35 |
+
|
36 |
+
# Prepare DNN input
|
37 |
+
Y = torch.unsqueeze(model._forward_transform(model._stft(y.to(args["device"]))), 0)
|
38 |
+
print(f"Transformed input in {time.time() - start_time:.2f}s")
|
39 |
+
|
40 |
+
Y = pad_spec(Y, mode="zero_pad") # Use "zero_pad" mode for padding
|
41 |
+
|
42 |
+
# Reverse sampling
|
43 |
+
sampler = model.get_pc_sampler(
|
44 |
+
'reverse_diffusion', args["corrector"], Y.to(args["device"]),
|
45 |
+
N=args["N"], corrector_steps=args["corrector_steps"], snr=args["snr"]
|
46 |
+
)
|
47 |
+
sample, _ = sampler()
|
48 |
+
|
49 |
+
# Backward transform in time domain
|
50 |
+
x_hat = model.to_audio(sample.squeeze(), T_orig)
|
51 |
+
|
52 |
+
# Renormalize
|
53 |
+
x_hat = x_hat * norm_factor
|
54 |
+
|
55 |
+
# Create a temporary path for saving the enhanced audio in Hugging Face Space
|
56 |
+
output_file = "/tmp/enhanced_output.wav" # Use a temporary directory
|
57 |
+
torchaudio.save(output_file, x_hat.cpu(), sr)
|
58 |
+
|
59 |
+
print(f"Processed audio in {time.time() - start_time:.2f}s")
|
60 |
+
|
61 |
+
# Return the path to the enhanced file for Gradio to handle
|
62 |
+
return output_file
|
63 |
+
|
64 |
+
# Gradio interface setup
|
65 |
+
inputs = gr.Audio(label="Input Audio", type="filepath") # Adjusted to 'filepath'
|
66 |
+
outputs = gr.Audio(label="Enhanced Audio", type="filepath") # Output as filepath
|
67 |
+
title = "Speech Enhancement using SGMSE"
|
68 |
+
description = "This Gradio demo uses the SGMSE model for speech enhancement. Upload your audio file to enhance it."
|
69 |
+
article = "<p style='text-align: center'><a href='https://huggingface.co/SP-UHH/speech-enhancement-sgmse' target='_blank'>Model Card</a></p>"
|
70 |
+
|
71 |
+
# Launch the Gradio interface
|
72 |
+
gr.Interface(fn=enhance_speech, inputs=inputs, outputs=outputs, title=title, description=description, article=article).launch()
|