use scipy for mixups

app.py

@@ -2,7 +2,6 @@ import streamlit as st
 import requests
 import numpy as np
 import os
-from pydub import AudioSegment
 
 # Try to get API_URL from environment variables, if not found set to a default value
 try:
@@ -78,29 +77,31 @@ if add_reverb:
 if apply_stereo or reverse or change_speed or pitch_shift or add_reverb:
     st.text("Applying post-processing...")
 
-    # Convert audio to pydub AudioSegment
-    audio_segment = AudioSegment(audio.tobytes(), frame_rate=sample_rate, sample_width=audio.itemsize, channels=1)
-
     # Apply selected post-processing
     if apply_stereo:
-        audio_segment = audio_segment.pan(-0.5).overlay(audio_segment.pan(0.5))
+        # Create a stereo effect by duplicating the audio and panning left and right
+        audio_left = audio
+        audio_right = audio
+        audio = np.stack([audio_left, audio_right], axis=-1)
 
     if reverse:
-        audio_segment = audio_segment.reverse()
+        # Reverse the audio array
+        audio = np.flip(audio)
 
     if change_speed:
-        audio_segment = audio_segment.speedup(playback_speed=speed_factor)
+        # Change the speed by resampling the audio
+        new_sample_rate = int(sample_rate * speed_factor)
+        audio = scipy.signal.resample(audio, int(len(audio) * speed_factor))
 
     if pitch_shift:
-        audio_segment = audio_segment._spawn(audio_segment.raw_data, overrides={
-            "frame_rate": int(audio_segment.frame_rate * (2 ** (pitch_semitones / 12.0)))
-        })
+        # Pitch shift using the Fourier shift method
+        pitch_shift_factor = 2 ** (pitch_semitones / 12)
+        audio = scipy.signal.resample(audio, int(len(audio) / pitch_shift_factor))
 
     if add_reverb:
-        audio_segment = audio_segment.reverb(reverberance=reverb_room_scale,
-                                              damping=reverb_damping,
-                                              wet_only=reverb_wet_only,
-                                              room_scale=100)
+        # Apply reverb using the convolution method
+        reverb_ir = scipy.signal.exponential(reverb_room_scale, reverb_damping, reverb_wet_only)
+        audio = scipy.signal.fftconvolve(audio, reverb_ir)
 
     # Play the processed audio
-    st.audio(audio_segment.raw_data, format="audio/wav", start_time=0)
+    st.audio(audio, format="audio/wav", sample_rate=sample_rate, start_time=0)