Spaces:
Running
Running
Commit
·
2a24191
1
Parent(s):
af4c524
Update Notebook, change resampling
Browse files
pvq_manipulation/Example_Notebook.ipynb
CHANGED
@@ -20,8 +20,9 @@
|
|
20 |
"from pvq_manipulation.models.ffjord import FFJORD\n",
|
21 |
"from IPython.display import display, Audio, clear_output\n",
|
22 |
"from pvq_manipulation.models.hubert import HubertExtractor, SID_LARGE_LAYER\n",
|
23 |
-
"from paderbox.transform.module_resample import resample_sox\n",
|
24 |
-
"
|
|
|
25 |
]
|
26 |
},
|
27 |
{
|
@@ -83,7 +84,7 @@
|
|
83 |
" model_name=\"HUBERT_LARGE\",\n",
|
84 |
" backend=\"torchaudio\",\n",
|
85 |
" device='cpu', \n",
|
86 |
-
" storage_dir
|
87 |
")"
|
88 |
]
|
89 |
},
|
@@ -151,7 +152,8 @@
|
|
151 |
"\n",
|
152 |
"def extract_speaker_embedding(example):\n",
|
153 |
" observation, sr = pb.io.load_audio(example['audio_path']['observation'], return_sample_rate=True)\n",
|
154 |
-
" observation = resample_sox(observation, in_rate=sr, out_rate=16_000)\n",
|
|
|
155 |
" \n",
|
156 |
" vad = EnergyVAD(sample_rate=16_000)\n",
|
157 |
" if observation.ndim == 1:\n",
|
@@ -206,16 +208,16 @@
|
|
206 |
},
|
207 |
{
|
208 |
"cell_type": "markdown",
|
209 |
-
"id": "
|
210 |
"metadata": {},
|
211 |
"source": [
|
212 |
-
"#
|
213 |
]
|
214 |
},
|
215 |
{
|
216 |
"cell_type": "code",
|
217 |
"execution_count": null,
|
218 |
-
"id": "
|
219 |
"metadata": {},
|
220 |
"outputs": [],
|
221 |
"source": [
|
@@ -239,6 +241,15 @@
|
|
239 |
")"
|
240 |
]
|
241 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
242 |
{
|
243 |
"cell_type": "code",
|
244 |
"execution_count": null,
|
@@ -252,11 +263,18 @@
|
|
252 |
" 'example_id': \"1034_121119_000028_000001\",\n",
|
253 |
"}\n",
|
254 |
"\n",
|
|
|
255 |
"label_options = ['Weight', 'Resonance', 'Breathiness', 'Roughness', 'Loudness', 'Strain', 'Pitch']\n",
|
256 |
"\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
257 |
"manipulation_idx_widget = widgets.Dropdown(\n",
|
258 |
" options=[(label, i) for i, label in enumerate(label_options)],\n",
|
259 |
-
" value=2,
|
260 |
" description='Type:',\n",
|
261 |
" style={'description_width': 'initial'}\n",
|
262 |
")\n",
|
@@ -294,9 +312,18 @@
|
|
294 |
" manipulation_fkt=manipulation_fkt,\n",
|
295 |
" )\n",
|
296 |
" \n",
|
|
|
|
|
|
|
|
|
|
|
297 |
" with audio_output:\n",
|
298 |
" clear_output(wait=True) \n",
|
|
|
299 |
" display(Audio(wav_manipulated, rate=24_000, normalize=True))\n",
|
|
|
|
|
|
|
300 |
" display(Audio(example['audio_path']['observation'], rate=24_000, normalize=True))\n",
|
301 |
"\n",
|
302 |
" print(f\"Manipulated {label_options[manipulation_idx]} with strength {manipulation_fkt}\")\n",
|
@@ -304,6 +331,22 @@
|
|
304 |
"run_button.on_click(update_manipulation)\n",
|
305 |
"display(manipulation_idx_widget, manipulation_fkt_widget, run_button, audio_output)"
|
306 |
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
307 |
}
|
308 |
],
|
309 |
"metadata": {
|
|
|
20 |
"from pvq_manipulation.models.ffjord import FFJORD\n",
|
21 |
"from IPython.display import display, Audio, clear_output\n",
|
22 |
"from pvq_manipulation.models.hubert import HubertExtractor, SID_LARGE_LAYER\n",
|
23 |
+
"#from paderbox.transform.module_resample import resample_sox # use resample_sox or librosa for resampling \n",
|
24 |
+
"import librosa\n",
|
25 |
+
"from pvq_manipulation.helper.vad import EnergyVAD"
|
26 |
]
|
27 |
},
|
28 |
{
|
|
|
84 |
" model_name=\"HUBERT_LARGE\",\n",
|
85 |
" backend=\"torchaudio\",\n",
|
86 |
" device='cpu', \n",
|
87 |
+
" storage_dir= # target storage dir hubert model\n",
|
88 |
")"
|
89 |
]
|
90 |
},
|
|
|
152 |
"\n",
|
153 |
"def extract_speaker_embedding(example):\n",
|
154 |
" observation, sr = pb.io.load_audio(example['audio_path']['observation'], return_sample_rate=True)\n",
|
155 |
+
" #observation = resample_sox(observation, in_rate=sr, out_rate=16_000)\n",
|
156 |
+
" observation = librosa.resample(observation, orig_sr=sr, target_sr=16_000)\n",
|
157 |
" \n",
|
158 |
" vad = EnergyVAD(sample_rate=16_000)\n",
|
159 |
" if observation.ndim == 1:\n",
|
|
|
208 |
},
|
209 |
{
|
210 |
"cell_type": "markdown",
|
211 |
+
"id": "941389a1-abd9-4cc6-80b3-5bbc7818e037",
|
212 |
"metadata": {},
|
213 |
"source": [
|
214 |
+
"# The following cell is intended for debugging purposes. If everything is functioning correctly, you can ignore it."
|
215 |
]
|
216 |
},
|
217 |
{
|
218 |
"cell_type": "code",
|
219 |
"execution_count": null,
|
220 |
+
"id": "14c279b3-6511-4a48-89c2-4efc613a08d3",
|
221 |
"metadata": {},
|
222 |
"outputs": [],
|
223 |
"source": [
|
|
|
241 |
")"
|
242 |
]
|
243 |
},
|
244 |
+
{
|
245 |
+
"cell_type": "markdown",
|
246 |
+
"id": "008035ba-6054-4e6e-ab16-1aaaf68f584a",
|
247 |
+
"metadata": {},
|
248 |
+
"source": [
|
249 |
+
"# Get example manipulation\n",
|
250 |
+
"### Manipulating a PVQ value: a_man = a + strength (where 'a' is the original)"
|
251 |
+
]
|
252 |
+
},
|
253 |
{
|
254 |
"cell_type": "code",
|
255 |
"execution_count": null,
|
|
|
263 |
" 'example_id': \"1034_121119_000028_000001\",\n",
|
264 |
"}\n",
|
265 |
"\n",
|
266 |
+
"labels = load_speaker_labels(example, config_norm_flow)\n",
|
267 |
"label_options = ['Weight', 'Resonance', 'Breathiness', 'Roughness', 'Loudness', 'Strain', 'Pitch']\n",
|
268 |
"\n",
|
269 |
+
"print('Estimated PVQ strengths of input speaker:')\n",
|
270 |
+
"max_len = max(len(name) for name in label_options) \n",
|
271 |
+
"for label_name, pvq in zip(label_options, labels):\n",
|
272 |
+
" print(f'{label_name:<{max_len}} : {pvq:6.2f}')\n",
|
273 |
+
"\n",
|
274 |
+
"\n",
|
275 |
"manipulation_idx_widget = widgets.Dropdown(\n",
|
276 |
" options=[(label, i) for i, label in enumerate(label_options)],\n",
|
277 |
+
" value=2, \n",
|
278 |
" description='Type:',\n",
|
279 |
" style={'description_width': 'initial'}\n",
|
280 |
")\n",
|
|
|
312 |
" manipulation_fkt=manipulation_fkt,\n",
|
313 |
" )\n",
|
314 |
" \n",
|
315 |
+
" wav_unmanipulated = tts_model.synthesize_from_example({\n",
|
316 |
+
" 'text': \"It took me quite a long time to develop a voice, and now that I have it I'm not going to be silent.\", \n",
|
317 |
+
" 'd_vector': d_vector.detach().numpy(),\n",
|
318 |
+
" })\n",
|
319 |
+
" \n",
|
320 |
" with audio_output:\n",
|
321 |
" clear_output(wait=True) \n",
|
322 |
+
" print('Manipulated Speaker')\n",
|
323 |
" display(Audio(wav_manipulated, rate=24_000, normalize=True))\n",
|
324 |
+
" print('Unmanipulated Synthese')\n",
|
325 |
+
" display(Audio(wav_unmanipulated, rate=24_000, normalize=True))\n",
|
326 |
+
" print('Original Speaker')\n",
|
327 |
" display(Audio(example['audio_path']['observation'], rate=24_000, normalize=True))\n",
|
328 |
"\n",
|
329 |
" print(f\"Manipulated {label_options[manipulation_idx]} with strength {manipulation_fkt}\")\n",
|
|
|
331 |
"run_button.on_click(update_manipulation)\n",
|
332 |
"display(manipulation_idx_widget, manipulation_fkt_widget, run_button, audio_output)"
|
333 |
]
|
334 |
+
},
|
335 |
+
{
|
336 |
+
"cell_type": "code",
|
337 |
+
"execution_count": null,
|
338 |
+
"id": "f551258d-df75-4d7d-a2bd-b9c52bbdf22b",
|
339 |
+
"metadata": {},
|
340 |
+
"outputs": [],
|
341 |
+
"source": []
|
342 |
+
},
|
343 |
+
{
|
344 |
+
"cell_type": "code",
|
345 |
+
"execution_count": null,
|
346 |
+
"id": "d185a73f-d066-465c-bcea-20ff6f279362",
|
347 |
+
"metadata": {},
|
348 |
+
"outputs": [],
|
349 |
+
"source": []
|
350 |
}
|
351 |
],
|
352 |
"metadata": {
|