update

app.py

@@ -51,10 +51,16 @@ def inference_gradio(input, model_choice):  # input is audio waveform in [T, cha
     spec_plot_gen = plot_spectrogram(spec_gen)
 
     output_audio = (h.sampling_rate, output) # tuple for gr.Audio output
-    output_image = PIL.Image.frombytes(
-        "RGB",
+
+    buffer = spec_plot_gen.canvas.buffer_rgba()
+    output_image = PIL.Image.frombuffer(
+        "RGBA",
         spec_plot_gen.canvas.get_width_height(),
-        spec_plot_gen.canvas.tostring_rgb(),
+        buffer,
+        "raw",
+        "RGBA",
+        0,
+        1
     )
 
     return output_audio, output_image