mixings with scipy

app.py

@@ -4,6 +4,11 @@ import numpy as np
 import os
 import base64
 from streamlit import session_state as st_state
+import librosa
+import soundfile as sf
+from scipy.signal import butter, sosfilt
+ 
+
 
 # Try to get API_URL from environment variables, if not found set to a default value
 try:
@@ -33,7 +38,15 @@ if 'augmented_audio' not in st_state:
 
 # Streamlit app title
 st.title("Songlabai")
- 
+
+# Initialize session state variables
+if 'vocal_audio' not in st_state:
+    st_state.vocal_audio = None
+if 'audio' not in st_state:
+    st_state.audio = None
+if 'augmented_audio' not in st_state:
+    st_state.augmented_audio = None
+    
 uploaded_file = st.file_uploader("Upload Music File", type=["mp3", "wav", "ogg", "flac", "aac"])
 
 genres = [
@@ -67,61 +80,107 @@ if st.button("Generate Audio"):
     sample_rate = response.json()[0]['sample_rate']
     st.audio(st_state.audio, format="audio/wav", sample_rate=sample_rate, start_time=0)
 
+ 
+vocal_file = st.file_uploader("Upload Vocal File", type=["mp3", "wav", "ogg", "flac", "aac"])
+
 # Post-processing options
 st.header("Post-processing Options")
-apply_stereo = st.checkbox("Apply Stereo Effect")
-reverse = st.checkbox("Reverse Audio")
-change_speed = st.checkbox("Change Speed")
-if change_speed:
-    speed_factor = st.slider("Speed Factor:", min_value=0.1, max_value=2.0, value=1.0, step=0.1)
-
-# Pitch shifting
-st.header("Pitch Shifting")
-pitch_shift = st.checkbox("Pitch Shift")
-if pitch_shift:
-    pitch_semitones = st.slider("Pitch (semitones):", min_value=-12, max_value=12, value=0, step=1)
+
+# Mixing
+mix_vocals = st.checkbox("Mix Vocals")
+if mix_vocals and st_state.vocal_audio is not None:
+    # Load the vocal audio
+    vocal_audio, _ = librosa.load(vocal_file, sr=sample_rate, mono=False)
+
+    # Adjust the vocal audio length to match the generated audio
+    vocal_audio = librosa.util.fix_length(vocal_audio, len(st_state.audio))
+
+    # Mix the vocal audio with the generated audio
+    st_state.augmented_audio = (st_state.audio + vocal_audio) / 2
+
+# Mastering
+st.subheader("Mastering")
+
+# Volume Balance
+volume_balance = st.slider("Volume Balance", min_value=-10.0, max_value=10.0, value=0.0, step=0.1)
+if st.button("Apply Volume Balance"):
+    if st_state.augmented_audio is None:
+        st_state.augmented_audio = st_state.audio
+    st_state.augmented_audio *= 10 ** (volume_balance / 20)
+
+# Compression
+compression_ratio = st.slider("Compression Ratio", min_value=1.0, max_value=10.0, value=3.0, step=0.1)
+if st.button("Apply Compression"):
+    if st_state.augmented_audio is None:
+        st_state.augmented_audio = st_state.audio
+
+    # Apply compression using a simple soft-knee compressor
+    threshold = -20  # dBFS
+    ratio = compression_ratio
+    knee = 10  # dB
+    max_gain = 20  # dB
+
+    def compress(x, threshold, ratio, knee, max_gain):
+        over = np.maximum(x - threshold, 0)
+        gain = over / (over + knee) * (1 - (1 / ratio)) + 1
+        gain = np.maximum(gain, 1 - max_gain)
+        return x * gain
+
+    st_state.augmented_audio = compress(st_state.augmented_audio, threshold, ratio, knee, max_gain)
+
+# EQ
+eq_low = st.slider("EQ Low", min_value=-10.0, max_value=10.0, value=0.0, step=0.1)
+eq_mid = st.slider("EQ Mid", min_value=-10.0, max_value=10.0, value=0.0, step=0.1)
+eq_high = st.slider("EQ High", min_value=-10.0, max_value=10.0, value=0.0, step=0.1)
+
+if st.button("Apply EQ"):
+    if st_state.augmented_audio is None:
+        st_state.augmented_audio = st_state.audio
+
+    # Apply a simple 3-band EQ using a butterworth filter
+    nyquist = sample_rate / 2
+    low_cutoff = 200 / nyquist
+    mid_cutoff = 2000 / nyquist
+    high_cutoff = 8000 / nyquist
+
+    low_sos = butter(4, low_cutoff, btype='low', output='sos', analog=False)
+    mid_sos = butter(4, [low_cutoff, mid_cutoff], btype='band', output='sos', analog=False)
+    high_sos = butter(4, high_cutoff, btype='high', output='sos', analog=False)
+
+    st_state.augmented_audio = sosfilt(np.dstack((low_sos, mid_sos, high_sos)),
+                                       st_state.augmented_audio,
+                                       np.stack((eq_low, eq_mid, eq_high)))
 
 # Reverb
-st.header("Reverb")
-add_reverb = st.checkbox("Add Reverb")
-if add_reverb:
-    reverb_room_scale = st.slider("Room Scale:", min_value=0.0, max_value=100.0, value=50.0)
-    reverb_damping = st.slider("Damping:", min_value=0.0, max_value=100.0, value=50.0)
-    reverb_wet_only = st.checkbox("Wet Only", value=False)
-
-# Apply selected post-processing
-if apply_stereo or reverse or change_speed or pitch_shift or add_reverb:
-    st.text("Applying post-processing...")
-
-    # Apply selected post-processing
-    st_state.augmented_audio = st_state.audio.copy()
-
-    if apply_stereo:
-        # Create a stereo effect by duplicating the audio and panning left and right
-        audio_left = st_state.augmented_audio
-        audio_right = st_state.augmented_audio
-        st_state.augmented_audio = np.stack([audio_left, audio_right], axis=-1)
-
-    if reverse:
-        # Reverse the audio array
-        st_state.augmented_audio = np.flip(st_state.augmented_audio)
-
-    if change_speed:
-        # Change the speed by resampling the audio
-        new_sample_rate = int(sample_rate * speed_factor)
-        st_state.augmented_audio = scipy.signal.resample(st_state.augmented_audio, int(len(st_state.augmented_audio) * speed_factor))
-
-    if pitch_shift:
-        # Pitch shift using the Fourier shift method
-        pitch_shift_factor = 2 ** (pitch_semitones / 12)
-        st_state.augmented_audio = scipy.signal.resample(st_state.augmented_audio, int(len(st_state.augmented_audio) / pitch_shift_factor))
-
-    if add_reverb:
-        # Apply reverb using the convolution method
-        reverb_ir = scipy.signal.exponential(reverb_room_scale, reverb_damping, reverb_wet_only)
-        st_state.augmented_audio = scipy.signal.fftconvolve(st_state.augmented_audio, reverb_ir)
-
-    # Play the processed audio
+reverb_amount = st.slider("Reverb Amount", min_value=0.0, max_value=1.0, value=0.5, step=0.01)
+if st.button("Apply Reverb"):
+    if st_state.augmented_audio is None:
+        st_state.augmented_audio = st_state.audio
+
+    # Apply a simple reverb effect using convolution
+    ir_length = int(sample_rate * 2.5)  # 2.5 seconds
+    ir = np.zeros(ir_length)
+    ir[0] = 1
+    ir = np.append(ir, np.zeros(len(st_state.augmented_audio) - ir_length))
+    reverb = np.convolve(st_state.augmented_audio, ir, mode='full')[:len(st_state.augmented_audio)]
+    st_state.augmented_audio = st_state.augmented_audio + reverb_amount * reverb
+
+# Delay
+delay_amount = st.slider("Delay Amount", min_value=0.0, max_value=1.0, value=0.5, step=0.01)
+delay_time = st.slider("Delay Time (ms)", min_value=10, max_value=500, value=100, step=10)
+if st.button("Apply Delay"):
+    if st_state.augmented_audio is None:
+        st_state.augmented_audio = st_state.audio
+
+    # Apply a simple delay effect
+    delay_samples = int(delay_time / 1000 * sample_rate)
+    delay = np.zeros(len(st_state.augmented_audio) + delay_samples)
+    delay[delay_samples:] = st_state.augmented_audio
+    delay[:len(st_state.augmented_audio)] += delay_amount * delay[:-delay_samples]
+    st_state.augmented_audio = delay[:len(st_state.augmented_audio)]
+
+# Display the final audio
+if st_state.augmented_audio is not None:
     st.audio(st_state.augmented_audio, format="audio/wav", sample_rate=sample_rate, start_time=0)
-
+    
 st.link_button("Download/Save", "https://songlabai.com/subcribe/")