Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -9,7 +9,8 @@ import numpy as np
|
|
| 9 |
import soundfile as sf
|
| 10 |
import os
|
| 11 |
import random
|
| 12 |
-
import tempfile
|
|
|
|
| 13 |
|
| 14 |
# Load your Pix2Pix model (make sure the path is correct)
|
| 15 |
model = load_model('./model_022600.h5', compile=False)
|
|
@@ -66,6 +67,20 @@ def modify_spectrogram(spectrogram):
|
|
| 66 |
|
| 67 |
return spectrogram
|
| 68 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
# Function to process the input image and convert to audio
|
| 70 |
def process_image(input_image):
|
| 71 |
# Load and preprocess the input image
|
|
@@ -93,6 +108,9 @@ def process_image(input_image):
|
|
| 93 |
# Modify the spectrogram randomly
|
| 94 |
img = modify_spectrogram(img)
|
| 95 |
|
|
|
|
|
|
|
|
|
|
| 96 |
# Convert the spectrogram back to audio using librosa
|
| 97 |
wav = librosa.feature.inverse.mel_to_audio(img, sr=44100, n_fft=2048, hop_length=512)
|
| 98 |
|
|
@@ -101,15 +119,15 @@ def process_image(input_image):
|
|
| 101 |
sf.write(temp_audio_file.name, wav, samplerate=44100)
|
| 102 |
audio_file_path = temp_audio_file.name
|
| 103 |
|
| 104 |
-
return audio_file_path # Return the
|
| 105 |
|
| 106 |
# Create a Gradio interface
|
| 107 |
interface = gr.Interface(
|
| 108 |
fn=process_image,
|
| 109 |
inputs=gr.Image(type="pil"), # Input is an image
|
| 110 |
-
outputs=gr.Audio(type="filepath"), # Output
|
| 111 |
-
title="Image to Audio Generator", # App title
|
| 112 |
-
description="Upload an image (preferably a spectrogram), and get an audio file generated using Pix2Pix.",
|
| 113 |
)
|
| 114 |
|
| 115 |
# Launch the interface
|
|
|
|
| 9 |
import soundfile as sf
|
| 10 |
import os
|
| 11 |
import random
|
| 12 |
+
import tempfile
|
| 13 |
+
import matplotlib.pyplot as plt
|
| 14 |
|
| 15 |
# Load your Pix2Pix model (make sure the path is correct)
|
| 16 |
model = load_model('./model_022600.h5', compile=False)
|
|
|
|
| 67 |
|
| 68 |
return spectrogram
|
| 69 |
|
| 70 |
+
# Function to save the modified spectrogram image for display
|
| 71 |
+
def save_spectrogram_image(spectrogram):
|
| 72 |
+
plt.figure(figsize=(10, 4))
|
| 73 |
+
plt.imshow(spectrogram, aspect='auto', origin='lower', cmap='gray')
|
| 74 |
+
plt.axis('off')
|
| 75 |
+
|
| 76 |
+
# Save to a temporary file
|
| 77 |
+
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as temp_image_file:
|
| 78 |
+
plt.savefig(temp_image_file.name, bbox_inches='tight', pad_inches=0)
|
| 79 |
+
temp_image_path = temp_image_file.name
|
| 80 |
+
|
| 81 |
+
plt.close()
|
| 82 |
+
return temp_image_path
|
| 83 |
+
|
| 84 |
# Function to process the input image and convert to audio
|
| 85 |
def process_image(input_image):
|
| 86 |
# Load and preprocess the input image
|
|
|
|
| 108 |
# Modify the spectrogram randomly
|
| 109 |
img = modify_spectrogram(img)
|
| 110 |
|
| 111 |
+
# Save the modified spectrogram as an image
|
| 112 |
+
spectrogram_image_path = save_spectrogram_image(img)
|
| 113 |
+
|
| 114 |
# Convert the spectrogram back to audio using librosa
|
| 115 |
wav = librosa.feature.inverse.mel_to_audio(img, sr=44100, n_fft=2048, hop_length=512)
|
| 116 |
|
|
|
|
| 119 |
sf.write(temp_audio_file.name, wav, samplerate=44100)
|
| 120 |
audio_file_path = temp_audio_file.name
|
| 121 |
|
| 122 |
+
return spectrogram_image_path, audio_file_path # Return the paths for both spectrogram image and audio
|
| 123 |
|
| 124 |
# Create a Gradio interface
|
| 125 |
interface = gr.Interface(
|
| 126 |
fn=process_image,
|
| 127 |
inputs=gr.Image(type="pil"), # Input is an image
|
| 128 |
+
outputs=[gr.Image(type="file"), gr.Audio(type="filepath")], # Output both spectrogram image and audio file
|
| 129 |
+
title="Image to Audio Generator with Spectrogram Display", # App title
|
| 130 |
+
description="Upload an image (preferably a spectrogram), and get an audio file generated using Pix2Pix. You can also see the modified spectrogram.",
|
| 131 |
)
|
| 132 |
|
| 133 |
# Launch the interface
|