file structuring

src/data_processing/audio_processing_validation.py

@@ -5,7 +5,7 @@
 
 import os
 import random
-from training_audio_information import compiled_data_path
+from usage.usage_specs import compiled_data_path
 from utils.signal_helpers import (
     stft_and_istft,
 )

src/data_processing/encode_audio_data.py

@@ -8,7 +8,7 @@
     load_loudness_data,
 )
 from utils.signal_helpers import encode_sample_directory
-from training_audio_information import training_audio_dir, compiled_data_path
+from usage.usage_specs import training_audio_dir, compiled_data_path
 
 # Encode audio samples
 if len(sys.argv) > 1:

src/stereo_sample_gan.py

@@ -16,7 +16,7 @@
     load_loudness_data,
 )
 
-from data_processing.training_audio_information import compiled_data_path
+from usage.usage_specs import compiled_data_path
 
 # Constants
 LR_G = 0.003

src/train.py

@@ -9,6 +9,7 @@
 )
 from utils.signal_helpers import graph_spectrogram
 
+
 # Constants
 N_EPOCHS = 8
 VALIDATION_INTERVAL = 4

src/usage/generate.py

@@ -0,0 +1,50 @@
+import sys
+import os
+
+sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+
+import torch
+from architecture import Generator, LATENT_DIM
+from usage.usage_specs import (
+    model_to_generate_with,
+    outputs_dir,
+    generated_audio_name,
+    audio_generation_count,
+    visualize_generated,
+)
+from utils.file_helpers import get_device, save_audio
+from utils.signal_helpers import audio_to_norm_db, graph_spectrogram, norm_db_to_audio
+
+# Initialize Generator
+device = get_device()
+
+generator = Generator()
+generator.load_state_dict(
+    torch.load(
+        model_to_generate_with, map_location=torch.device(device), weights_only=False
+    )
+)
+generator.eval()
+
+# Generate audio
+z = torch.randn(audio_generation_count, LATENT_DIM, 1, 1)
+with torch.no_grad():
+    generated_output = generator(z)
+
+
+generated_output = generated_output.squeeze().numpy()
+print("Generated output shape:", generated_output.shape)
+
+# Visualize and save audio
+for i in range(audio_generation_count):
+    current_sample = generated_output[i]
+
+    audio_info = norm_db_to_audio(current_sample)
+    audio_save_path = os.path.join(outputs_dir, f"{generated_audio_name}-{i + 1}.wav")
+
+    save_audio(audio_save_path, audio_info)
+
+    if visualize_generated is True:
+        vis_signal_after_istft = audio_to_norm_db(audio_info)
+        graph_spectrogram(vis_signal_after_istft, "generated audio (after istft)")

src/usage/usage_specs.py

@@ -0,0 +1,16 @@
+# Processing training data
+training_audio_dir = "data/one_shots"  # Your training data path
+compiled_data_path = "data/compiled_data.npy"  # Your compiled data/output path
+training_sample_length = 0.6  # seconds
+
+# Saving model
+outputs_dir = "outputs"  # Where to save your generated audio & model
+model_save_name = "StereoSampleGAN-OldKick"  # What to name your model save
+model_save_path = f"{outputs_dir}/{model_save_name}.pth"
+
+# Generating audio
+model_to_generate_with = model_save_path  # Generation model path
+audio_generation_count = 2  # Audio examples to generate
+generated_audio_name = "generated_audio"  # Output file name
+generated_sample_length = 0.6  # Match model training data audio length
+visualize_generated = True  # SHow generated audio spectrogra,s

src/utils/file_helpers.py

@@ -3,8 +3,9 @@
 import torch
 import soundfile as sf
 
+from usage.usage_specs import model_save_path
+
 # Constants
-outputs_dir = "outputs"
 GLOBAL_SR = 44100
 
 
@@ -25,7 +26,7 @@ def save_model(model, name):
     # Save model
     torch.save(
         model.state_dict(),
-        f"{outputs_dir}/{name}.pth",
+        model_save_path,
     )
     print(f"Model Saved")
 

src/utils/signal_helpers.py

@@ -10,10 +10,12 @@
 import plotly.subplots as sp
 import scipy
 
-from data_processing.training_audio_information import audio_sample_length
+from usage.usage_specs import (
+    training_sample_length,
+    outputs_dir,
+)
 from utils.file_helpers import (
     GLOBAL_SR,
-    outputs_dir,
     delete_DSStore,
     save_audio,
     save_loudness_data,
@@ -25,7 +27,7 @@
 
 # STFT Helpers
 GLOBAL_WIN = 510
-GLOBAL_HOP = int(audio_sample_length * GLOBAL_SR) // (DATA_SHAPE - 1)
+GLOBAL_HOP = int(training_sample_length * GLOBAL_SR) // (DATA_SHAPE - 1)
 window = scipy.signal.windows.kaiser(GLOBAL_WIN, beta=12)
 
 
@@ -85,7 +87,7 @@ def griffin_lim_istft(channel_magnitudes):
             center=True,
         )
         y = librosa.util.fix_length(
-            y, size=int(audio_sample_length * GLOBAL_SR), axis=0
+            y, size=int(training_sample_length * GLOBAL_SR), axis=0
         )
 
         if i > 0:
@@ -124,7 +126,7 @@ def load_audio(path):
     y, sr = librosa.load(path, sr=GLOBAL_SR, mono=False)
     if y.ndim == 1:
         y = np.stack((y, y), axis=0)
-    y = librosa.util.fix_length(y, size=int(audio_sample_length * GLOBAL_SR), axis=1)
+    y = librosa.util.fix_length(y, size=int(training_sample_length * GLOBAL_SR), axis=1)
     return y
 
 
@@ -194,10 +196,10 @@ def graph_spectrogram(audio_data, sample_name):
 def generate_sine_impulses(num_impulses=1, outPath="model"):
     amplitude = 1
     for i in range(num_impulses):
-        t = np.arange(0, audio_sample_length, 1 / GLOBAL_SR)
+        t = np.arange(0, training_sample_length, 1 / GLOBAL_SR)
         freq = np.random.uniform(0, 20000)
         audio_wave = amplitude * np.sin(2 * np.pi * freq * t)
-        num_samples = int(audio_sample_length * GLOBAL_SR)
+        num_samples = int(training_sample_length * GLOBAL_SR)
         audio_signal = np.zeros(num_samples)
 
         audio_wave = audio_wave[:num_samples]