eighth commit

.gitignore

@@ -0,0 +1 @@
+.vscode*

app.py

@@ -44,6 +44,33 @@ def separate_track_with_model(
         return estimates
 
 
+def parallel_mix(input, output, mix_coefficient):
+    sr = 44100
+    return sr, input[1] * mix_coefficient + output[1] * (1 - mix_coefficient)
+
+
+def int16_to_float32(wav):
+    X = wav / 32768
+    return X
+
+
+def waveform_plot(input, output, prl_mix_ouptut, figsize_x=20, figsize_y=9):
+    sr = 44100
+    fig, ax = plt.subplots(
+        nrows=3, sharex=True, sharey=True, figsize=(figsize_x, figsize_y)
+    )
+    librosa.display.waveshow(int16_to_float32(input[1]).T, sr=sr, ax=ax[0])
+    ax[0].set(title="Loudness Normalized Input")
+    ax[0].label_outer()
+    librosa.display.waveshow(int16_to_float32(output[1]).T, sr=sr, ax=ax[1])
+    ax[1].set(title="De-limiter Output")
+    ax[1].label_outer()
+    librosa.display.waveshow(int16_to_float32(prl_mix_ouptut[1]).T, sr=sr, ax=ax[2])
+    ax[2].set(title="Parallel Mix of the Input and its De-limiter Output")
+    ax[2].label_outer()
+    return fig
+
+
 def main(input, mix_coefficient):
     parser = argparse.ArgumentParser(description="model test.py")
     parser.add_argument("--target", type=str, default="all")
@@ -143,45 +170,18 @@ def main(input, mix_coefficient):
     )
 
     # if args.save_mixed_output:
-    track_lufs = meter.integrated_loudness(orig_audio)
+    orig_audio = int16_to_float32(orig_audio)
+    track_lufs = meter.integrated_loudness(orig_audio.T)
     augmented_gain = args.save_output_loudnorm - track_lufs
     orig_audio = orig_audio * db2linear(augmented_gain, eps=0.0)
 
     return (
         (sr, estimates.T),
-        (sr, orig_audio),
-        (sr, orig_audio * mix_coefficient + estimates.T * (1 - mix_coefficient)),
+        (sr, orig_audio.T),
+        (sr, orig_audio.T * mix_coefficient + estimates.T * (1 - mix_coefficient)),
     )
 
 
-def parallel_mix(input, output, mix_coefficient):
-    sr = 44100
-    return sr, input[1] * mix_coefficient + output[1] * (1 - mix_coefficient)
-
-
-def int16_to_float32(wav):
-    wav = np.frombuffer(wav, dtype=np.int16)
-    X = wav / 32768
-    return X
-
-
-def waveform_plot(input, output, prl_mix_ouptut, figsize_x=20, figsize_y=9):
-    sr = 44100
-    fig, ax = plt.subplots(
-        nrows=3, sharex=True, sharey=True, figsize=(figsize_x, figsize_y)
-    )
-    librosa.display.waveshow(int16_to_float32(input[1]).T, sr=sr, ax=ax[0])
-    ax[0].set(title="Loudness Normalized Input")
-    ax[0].label_outer()
-    librosa.display.waveshow(int16_to_float32(output[1]).T, sr=sr, ax=ax[1])
-    ax[1].set(title="De-limiter Output")
-    ax[1].label_outer()
-    librosa.display.waveshow(int16_to_float32(prl_mix_ouptut[1]).T, sr=sr, ax=ax[2])
-    ax[2].set(title="Parallel Mix of the Input and its De-limiter Output")
-    ax[2].label_outer()
-    return fig
-
-
 with gr.Blocks() as demo:
     gr.HTML(
         """
@@ -203,7 +203,7 @@ with gr.Blocks() as demo:
                 You can first upload a music (.wav or .mp3) file and then press "De-limit" button to apply the De-limiter. Since we use a CPU instead of a GPU, it may require a few minute.
                 Then, you can apply a Parallel Mix technique, which is a simple linear mixing technique of "loudness normalized input" and the "de-limiter output".
                 You can modify the mixing coefficient by yourself.
-                If the coefficient is 0.3 then the output will be the "loudness_normalized_input * 0.3 + de-limiter_output * 0.7"
+                If the coefficient is 0.3 then the output will be "loudness_normalized_input * 0.3 + de-limiter_output * 0.7"
             </div>
         """
     )