From e76132320e5ccff14a60d4da415dad8ebc3f0aa4 Mon Sep 17 00:00:00 2001
From: aryanKaga <162778626+aryanKaga@users.noreply.github.com>
Date: Wed, 23 Jul 2025 12:49:48 +0530
Subject: [PATCH] Add files via upload

---
 create_dataset.py | 80 +++++++++++++++++++++++++++++++++++++++++++++++
 stft.py           | 34 ++++++++++++++++++++
 2 files changed, 114 insertions(+)
 create mode 100644 create_dataset.py
 create mode 100644 stft.py

diff --git a/create_dataset.py b/create_dataset.py
new file mode 100644
index 00000000..b62a465a
--- /dev/null
+++ b/create_dataset.py
@@ -0,0 +1,80 @@
+import os
+import numpy as np
+import threading
+import time
+import psutil
+from stft import convert_to_magnitude,get_duration_difference_ms
+import re
+
+
+dataset={"input":[],"output":[]}
+
+
+
+
+terminal_direc=os.getcwd()
+
+clean_speech_dir=os.path.join(terminal_direc,'CleanSpeech_training') # directory for clean files
+noisy_speech_dir=os.path.join(terminal_direc,'data10','NoisySpeech_training') #directory for noisy files
+
+clean_speech=os.listdir(clean_speech_dir)
+noisy_speech=os.listdir(noisy_speech_dir)
+
+total_length_list=len(os.listdir(clean_speech_dir)) # the total length depends on unique clean speech available as data
+
+dataset_list = [{"input": [], "output": []} for _ in range(total_length_list+1)]
+
+for filename in clean_speech:
+    file_digit = int(re.search(r'clnsp(\d+)\.wav',filename).group(1))
+   
+    dataset_list[file_digit]['output'].append(filename)
+
+
+maxindex=0
+for filename in noisy_speech:
+    file_digit = int(re.search(r'clnsp(\d+)\.wav',filename).group(1))
+    
+    dataset_list[file_digit]['input'].append(filename)
+
+
+
+def create_mag_spectra_dataset():
+    
+    mag_data={"input":[],"output":[]}
+
+
+    for data in range(len(dataset_list)):
+
+        input_file=data['input'] #its the clean file to be extracted
+
+        input_mag=convert_to_magnitude(os.path.join(clean_speech_dir,input_file))
+
+        output_file=data['output'] # its list of noisy file corresponding to the above clean file
+
+
+        for filename in output_file:
+
+            output_mag=convert_to_magnitude(os.path.join(noisy_speech_dir,filename))
+
+            min_width=min(input_mag.shape[1],output_mag.shape[1])
+
+            input_mag=input_mag[:,:min_width]
+
+            output_mag=output_mag[:,:min_width]
+
+            mag_data['input'].append(input_mag)
+            mag_data['output'].append(output_mag)
+
+        
+    return mag_data
+            
+            
+            
+            
+
+
+
+
+
+
+
diff --git a/stft.py b/stft.py
new file mode 100644
index 00000000..a8b6802c
--- /dev/null
+++ b/stft.py
@@ -0,0 +1,34 @@
+import librosa
+import soundfile
+import os
+import numpy as np
+#convert a soundfile to stft and return 
+
+
+
+working_direc=os.getcwd()
+def convert_to_stft(path):
+    amp,sr=librosa.load(path,sr=16000)
+    return librosa.stft(amp)
+
+
+def convert_to_magnitude(path):
+    amp,sr=librosa.load(path,sr=16000)
+    return np.abs(librosa.stft(amp))
+
+
+def get_duration_difference_ms(file1, file2, sr=16000):
+    y1, _ = librosa.load(file1, sr=sr)
+    y2, _ = librosa.load(file2, sr=sr)
+
+    len1 = len(y1)
+    len2 = len(y2)
+
+    diff_samples = abs(len1 - len2)
+    diff_ms = (diff_samples / sr) * 1000
+
+    print(f"Length 1: {len1} samples ({len1/sr:.3f} s)")
+    print(f"Length 2: {len2} samples ({len2/sr:.3f} s)")
+    print(f"Difference: {diff_samples} samples = {diff_ms:.2f} ms")
+
+    return diff_ms
\ No newline at end of file