fix steps=None

audiodiffusion/__init__.py

@@ -206,7 +206,9 @@ class AudioDiffusionPipeline(DiffusionPipeline):
             if start_step > 0:
                 images[0, 0] = self.scheduler.add_noise(
                     torch.tensor(input_images[:, np.newaxis, np.newaxis, :]),
-                    noise, torch.tensor(steps - start_step))
+                    noise,
+                    torch.tensor(self.scheduler.num_train_timesteps -
+                                 start_step))
 
             pixels_per_second = (mel.get_sample_rate() / mel.hop_length)
             mask_start = int(mask_start_secs * pixels_per_second)

notebooks/test_model.ipynb

@@ -136,21 +136,6 @@
     "Try playing around with `start_steps`. Values closer to zero will produce new samples, while values closer to 1,000 will produce samples more faithful to the original."
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "a7e637e5",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "seed = 16183389798189209330  #@param {type:\"integer\"}\n",
-    "generator.manual_seed(seed)\n",
-    "image, (sample_rate, audio) = audio_diffusion.generate_spectrogram_and_audio(\n",
-    "    generator=generator)\n",
-    "display(image)\n",
-    "display(Audio(audio, rate=sample_rate))"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,