Spaces:
Running
Running
jackvinati
commited on
Commit
•
141afc1
1
Parent(s):
912effa
wavewizard_app.py
Browse files
app.py
ADDED
@@ -0,0 +1,198 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import librosa
|
3 |
+
import librosa.display
|
4 |
+
import matplotlib.pyplot as plt
|
5 |
+
import soundfile as sf
|
6 |
+
import gradio as gr
|
7 |
+
import io
|
8 |
+
import os
|
9 |
+
import base64
|
10 |
+
|
11 |
+
|
12 |
+
def analyze_audio_files(files, folder_path):
|
13 |
+
output_html = ""
|
14 |
+
file_paths = []
|
15 |
+
|
16 |
+
# Handle inputs: files can be a list of file paths or a folder path
|
17 |
+
if files:
|
18 |
+
file_paths.extend(files)
|
19 |
+
if folder_path:
|
20 |
+
if os.path.isdir(folder_path):
|
21 |
+
folder_files = [os.path.join(folder_path, f) for f in os.listdir(folder_path)
|
22 |
+
if os.path.isfile(os.path.join(folder_path, f))]
|
23 |
+
file_paths.extend(folder_files)
|
24 |
+
else:
|
25 |
+
return f"<p><strong>Folder not found:</strong> {folder_path}</p>"
|
26 |
+
|
27 |
+
for audio_file in file_paths:
|
28 |
+
try:
|
29 |
+
# Load the audio file
|
30 |
+
y, sr = librosa.load(audio_file, sr=None)
|
31 |
+
|
32 |
+
# Get original bit depth from file metadata
|
33 |
+
with sf.SoundFile(audio_file) as f:
|
34 |
+
bit_depth_info = f.subtype_info
|
35 |
+
|
36 |
+
# Time domain analysis
|
37 |
+
duration = len(y) / sr
|
38 |
+
|
39 |
+
# Frequency domain analysis
|
40 |
+
desired_freq_resolution = 10.0 # in Hz
|
41 |
+
|
42 |
+
# Calculate n_fft, limit it to a reasonable range
|
43 |
+
n_fft = int(sr / desired_freq_resolution)
|
44 |
+
n_fft = 2 ** int(np.ceil(np.log2(n_fft))) # Next power of two
|
45 |
+
|
46 |
+
# Set maximum and minimum n_fft to avoid excessive computation
|
47 |
+
max_n_fft = 32768
|
48 |
+
min_n_fft = 1024
|
49 |
+
n_fft = min(max(n_fft, min_n_fft), max_n_fft)
|
50 |
+
|
51 |
+
hop_length = n_fft // 4
|
52 |
+
|
53 |
+
# Compute the Short-Time Fourier Transform (STFT)
|
54 |
+
S = np.abs(librosa.stft(y, n_fft=n_fft, hop_length=hop_length))
|
55 |
+
|
56 |
+
# Compute the spectrogram (in dB)
|
57 |
+
S_db = librosa.amplitude_to_db(S, ref=np.max)
|
58 |
+
|
59 |
+
# Average over time to get the frequency spectrum
|
60 |
+
S_mean = np.mean(S, axis=1)
|
61 |
+
freqs = np.linspace(0, sr / 2, len(S_mean))
|
62 |
+
|
63 |
+
# Plot the frequency spectrum
|
64 |
+
fig1 = plt.figure(figsize=(8, 4))
|
65 |
+
plt.semilogx(freqs, 20 * np.log10(S_mean + 1e-10)) # Avoid log(0)
|
66 |
+
plt.xlabel('Frequency (Hz)', fontsize=12)
|
67 |
+
plt.ylabel('Amplitude (dB)', fontsize=12)
|
68 |
+
plt.title('Frequency Spectrum', fontsize=14)
|
69 |
+
plt.grid(True, which='both', ls='--')
|
70 |
+
plt.xlim(20, sr / 2)
|
71 |
+
plt.tight_layout()
|
72 |
+
spectrum_image = io.BytesIO()
|
73 |
+
plt.savefig(spectrum_image, format='png', bbox_inches='tight')
|
74 |
+
plt.close(fig1)
|
75 |
+
spectrum_image.seek(0)
|
76 |
+
spectrum_base64 = base64.b64encode(
|
77 |
+
spectrum_image.read()).decode('utf-8')
|
78 |
+
spectrum_html = f'<img src="data:image/png;base64,{spectrum_base64}" alt="Frequency Spectrum">'
|
79 |
+
|
80 |
+
# Plot the spectrogram
|
81 |
+
fig3 = plt.figure(figsize=(8, 4))
|
82 |
+
librosa.display.specshow(
|
83 |
+
S_db, sr=sr, x_axis='time', y_axis='linear', hop_length=hop_length)
|
84 |
+
plt.colorbar(format='%+2.0f dB')
|
85 |
+
plt.title('Spectrogram', fontsize=14)
|
86 |
+
plt.xlabel('Time (s)', fontsize=12)
|
87 |
+
plt.ylabel('Frequency (Hz)', fontsize=12)
|
88 |
+
plt.tight_layout()
|
89 |
+
spectrogram_image = io.BytesIO()
|
90 |
+
plt.savefig(spectrogram_image, format='png', bbox_inches='tight')
|
91 |
+
plt.close(fig3)
|
92 |
+
spectrogram_image.seek(0)
|
93 |
+
spectrogram_base64 = base64.b64encode(
|
94 |
+
spectrogram_image.read()).decode('utf-8')
|
95 |
+
spectrogram_html = f'<img src="data:image/png;base64,{spectrogram_base64}" alt="Spectrogram">'
|
96 |
+
|
97 |
+
# Analyze high-frequency content
|
98 |
+
# Define a threshold relative to the maximum amplitude
|
99 |
+
threshold_db = -80 # dB
|
100 |
+
max_amplitude_db = 20 * np.log10(np.max(S_mean + 1e-10))
|
101 |
+
threshold_amplitude_db = max_amplitude_db + threshold_db
|
102 |
+
threshold_amplitude = 10 ** (threshold_amplitude_db / 20)
|
103 |
+
|
104 |
+
# Find the highest frequency with significant content
|
105 |
+
significant_indices = np.where(S_mean >= threshold_amplitude)[0]
|
106 |
+
if len(significant_indices) > 0:
|
107 |
+
highest_freq = freqs[significant_indices[-1]]
|
108 |
+
|
109 |
+
# Estimate the real sample rate
|
110 |
+
estimated_sample_rate = highest_freq * 2 # Nyquist theorem
|
111 |
+
|
112 |
+
significant_freq_text = f"{highest_freq:.2f} Hz"
|
113 |
+
estimated_sample_rate_text = f"{estimated_sample_rate / 1000:.2f} kHz"
|
114 |
+
else:
|
115 |
+
significant_freq_text = "No significant frequency content detected."
|
116 |
+
estimated_sample_rate_text = "N/A"
|
117 |
+
|
118 |
+
# Estimate effective bit depth
|
119 |
+
# Calculate the signal's dynamic range
|
120 |
+
signal_rms = np.sqrt(np.mean(y ** 2))
|
121 |
+
noise_floor = np.percentile(np.abs(y), 0.1)
|
122 |
+
# Avoid division by zero
|
123 |
+
dynamic_range_db = 20 * \
|
124 |
+
np.log10(signal_rms / (noise_floor + 1e-10))
|
125 |
+
|
126 |
+
estimated_bit_depth = int(np.ceil(dynamic_range_db / 6.02))
|
127 |
+
|
128 |
+
# Prepare the output text as an HTML table
|
129 |
+
output_text = f"""
|
130 |
+
<h3 style="font-size:22px;">{os.path.basename(audio_file)}</h3>
|
131 |
+
<table style="font-size:18px;">
|
132 |
+
<tr><td><strong>File Bit Depth:</strong></td><td>{bit_depth_info}</td></tr>
|
133 |
+
<tr><td><strong>Sample Rate:</strong></td><td>{sr} Hz</td></tr>
|
134 |
+
<tr><td><strong>Duration:</strong></td><td>{duration:.2f} seconds</td></tr>
|
135 |
+
<tr><td><strong>Using n_fft =</strong></td><td>{n_fft}</td></tr>
|
136 |
+
<tr><td><strong>Significant frequency content up to:</strong></td><td>{significant_freq_text}</td></tr>
|
137 |
+
<tr><td><strong>Estimated Real Sample Rate:</strong></td><td>{estimated_sample_rate_text}</td></tr>
|
138 |
+
<tr><td><strong>Estimated Dynamic Range:</strong></td><td>{dynamic_range_db:.2f} dB</td></tr>
|
139 |
+
<tr><td><strong>Estimated Effective Bit Depth:</strong></td><td>{estimated_bit_depth} bits PCM</td></tr>
|
140 |
+
</table>
|
141 |
+
"""
|
142 |
+
|
143 |
+
# Plot histogram of sample values
|
144 |
+
fig2 = plt.figure(figsize=(8, 4))
|
145 |
+
plt.hist(y, bins=1000, alpha=0.7, color='blue',
|
146 |
+
edgecolor='black', log=True)
|
147 |
+
plt.xlabel('Amplitude', fontsize=12)
|
148 |
+
plt.ylabel('Count (log scale)', fontsize=12)
|
149 |
+
plt.title('Histogram of Sample Amplitudes', fontsize=14)
|
150 |
+
plt.grid(True)
|
151 |
+
plt.tight_layout()
|
152 |
+
histogram_image = io.BytesIO()
|
153 |
+
plt.savefig(histogram_image, format='png', bbox_inches='tight')
|
154 |
+
plt.close(fig2)
|
155 |
+
histogram_image.seek(0)
|
156 |
+
histogram_base64 = base64.b64encode(
|
157 |
+
histogram_image.read()).decode('utf-8')
|
158 |
+
histogram_html = f'<img src="data:image/png;base64,{histogram_base64}" alt="Histogram of Sample Amplitudes">'
|
159 |
+
|
160 |
+
# Combine text and images into HTML
|
161 |
+
output_html += f"""
|
162 |
+
{output_text}
|
163 |
+
<h4 style="font-size:20px;">Frequency Spectrum</h4>
|
164 |
+
{spectrum_html}
|
165 |
+
<h4 style="font-size:20px;">Spectrogram</h4>
|
166 |
+
{spectrogram_html}
|
167 |
+
<h4 style="font-size:20px;">Histogram of Sample Amplitudes</h4>
|
168 |
+
{histogram_html}
|
169 |
+
<hr>
|
170 |
+
"""
|
171 |
+
except Exception as e:
|
172 |
+
# Handle errors gracefully
|
173 |
+
output_html += f"<p><strong>File:</strong> {os.path.basename(audio_file)}</p><p><strong>Error:</strong> {str(e)}</p><hr>"
|
174 |
+
|
175 |
+
# Return the aggregated HTML output
|
176 |
+
return output_html
|
177 |
+
|
178 |
+
|
179 |
+
with gr.Blocks() as demo:
|
180 |
+
gr.Markdown("Wave Wizard")
|
181 |
+
gr.Markdown(
|
182 |
+
"Upload one or more audio files, or specify a folder containing audio files.")
|
183 |
+
with gr.Row():
|
184 |
+
file_input = gr.Files(label="Upload Audio Files",
|
185 |
+
type="filepath", file_count="multiple")
|
186 |
+
folder_input = gr.Textbox(label="Folder Path (optional)",
|
187 |
+
placeholder="Enter folder path containing audio files")
|
188 |
+
analyze_button = gr.Button("Analyze")
|
189 |
+
output_display = gr.HTML()
|
190 |
+
|
191 |
+
def analyze_wrapper(files, folder_path):
|
192 |
+
outputs = analyze_audio_files(files, folder_path)
|
193 |
+
return outputs
|
194 |
+
|
195 |
+
analyze_button.click(analyze_wrapper, inputs=[
|
196 |
+
file_input, folder_input], outputs=output_display)
|
197 |
+
|
198 |
+
demo.launch()
|