Spaces:
Running
on
T4
Running
on
T4
Update app.py
Browse files
app.py
CHANGED
@@ -253,14 +253,16 @@ def predict(
|
|
253 |
|
254 |
wav_chunks = []
|
255 |
## Direct mode
|
256 |
-
|
257 |
print("I: Generating new audio...")
|
258 |
t0 = time.time()
|
259 |
out = model.inference(
|
260 |
prompt,
|
261 |
language,
|
262 |
gpt_cond_latent,
|
263 |
-
speaker_embedding
|
|
|
|
|
264 |
)
|
265 |
inference_time = time.time() - t0
|
266 |
print(f"I: Time to generate audio: {round(inference_time*1000)} milliseconds")
|
@@ -269,8 +271,9 @@ def predict(
|
|
269 |
print(f"Real-time factor (RTF): {real_time_factor}")
|
270 |
metrics_text+=f"Real-time factor (RTF): {real_time_factor:.2f}\n"
|
271 |
torchaudio.save("output.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000)
|
272 |
-
"""
|
273 |
|
|
|
|
|
274 |
print("I: Generating new audio in streaming mode...")
|
275 |
t0 = time.time()
|
276 |
chunks = model.inference_stream(
|
@@ -305,6 +308,7 @@ def predict(
|
|
305 |
metrics_text += f"Real-time factor (RTF): {real_time_factor:.2f}\n"
|
306 |
|
307 |
torchaudio.save("output.wav", wav.squeeze().unsqueeze(0).cpu(), 24000)
|
|
|
308 |
|
309 |
except RuntimeError as e:
|
310 |
if "device-side assert" in str(e):
|
|
|
253 |
|
254 |
wav_chunks = []
|
255 |
## Direct mode
|
256 |
+
|
257 |
print("I: Generating new audio...")
|
258 |
t0 = time.time()
|
259 |
out = model.inference(
|
260 |
prompt,
|
261 |
language,
|
262 |
gpt_cond_latent,
|
263 |
+
speaker_embedding,
|
264 |
+
repetition_penalty=5.0,
|
265 |
+
temperature=0.75,
|
266 |
)
|
267 |
inference_time = time.time() - t0
|
268 |
print(f"I: Time to generate audio: {round(inference_time*1000)} milliseconds")
|
|
|
271 |
print(f"Real-time factor (RTF): {real_time_factor}")
|
272 |
metrics_text+=f"Real-time factor (RTF): {real_time_factor:.2f}\n"
|
273 |
torchaudio.save("output.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000)
|
|
|
274 |
|
275 |
+
|
276 |
+
"""
|
277 |
print("I: Generating new audio in streaming mode...")
|
278 |
t0 = time.time()
|
279 |
chunks = model.inference_stream(
|
|
|
308 |
metrics_text += f"Real-time factor (RTF): {real_time_factor:.2f}\n"
|
309 |
|
310 |
torchaudio.save("output.wav", wav.squeeze().unsqueeze(0).cpu(), 24000)
|
311 |
+
"""
|
312 |
|
313 |
except RuntimeError as e:
|
314 |
if "device-side assert" in str(e):
|