Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -69,86 +69,148 @@ class VibeVoiceDemo:
|
|
69 |
return np.array([])
|
70 |
|
71 |
@GPU
|
72 |
-
def generate_podcast(self,
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
81 |
|
82 |
-
|
83 |
-
|
84 |
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
|
|
89 |
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
108 |
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
|
|
117 |
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
|
|
|
|
|
|
127 |
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
if audio.ndim > 1:
|
136 |
-
audio = audio.squeeze()
|
137 |
|
138 |
-
|
|
|
|
|
|
|
|
|
139 |
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
|
|
|
|
|
|
146 |
|
147 |
-
|
148 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
149 |
|
150 |
-
|
151 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
152 |
|
153 |
|
154 |
|
|
|
69 |
return np.array([])
|
70 |
|
71 |
@GPU
|
72 |
+
def generate_podcast(self,
|
73 |
+
num_speakers: int,
|
74 |
+
script: str,
|
75 |
+
speaker_1: str = None,
|
76 |
+
speaker_2: str = None,
|
77 |
+
speaker_3: str = None,
|
78 |
+
speaker_4: str = None,
|
79 |
+
cfg_scale: float = 1.3):
|
80 |
+
"""
|
81 |
+
Generates a podcast as a single audio file from a script and saves it.
|
82 |
+
This is a non-streaming function.
|
83 |
+
"""
|
84 |
+
try:
|
85 |
+
# 1. Set generating state and validate inputs
|
86 |
+
self.is_generating = True
|
87 |
+
|
88 |
+
if not script.strip():
|
89 |
+
raise gr.Error("Error: Please provide a script.")
|
90 |
+
|
91 |
+
# Defend against common mistake with apostrophes
|
92 |
+
script = script.replace("β", "'")
|
93 |
|
94 |
+
if not 1 <= num_speakers <= 4:
|
95 |
+
raise gr.Error("Error: Number of speakers must be between 1 and 4.")
|
96 |
|
97 |
+
# 2. Collect and validate selected speakers
|
98 |
+
selected_speakers = [speaker_1, speaker_2, speaker_3, speaker_4][:num_speakers]
|
99 |
+
for i, speaker_name in enumerate(selected_speakers):
|
100 |
+
if not speaker_name or speaker_name not in self.available_voices:
|
101 |
+
raise gr.Error(f"Error: Please select a valid speaker for Speaker {i+1}.")
|
102 |
|
103 |
+
# 3. Build initial log
|
104 |
+
log = f"ποΈ Generating podcast with {num_speakers} speakers\n"
|
105 |
+
log += f"π Parameters: CFG Scale={cfg_scale}\n"
|
106 |
+
log += f"π Speakers: {', '.join(selected_speakers)}\n"
|
107 |
+
|
108 |
+
# 4. Load voice samples
|
109 |
+
voice_samples = []
|
110 |
+
for speaker_name in selected_speakers:
|
111 |
+
audio_path = self.available_voices[speaker_name]
|
112 |
+
# Assuming self.read_audio is a method in your class that returns audio data
|
113 |
+
audio_data = self.read_audio(audio_path)
|
114 |
+
if len(audio_data) == 0:
|
115 |
+
raise gr.Error(f"Error: Failed to load audio for {speaker_name}")
|
116 |
+
voice_samples.append(audio_data)
|
117 |
+
|
118 |
+
log += f"β
Loaded {len(voice_samples)} voice samples\n"
|
119 |
|
120 |
+
# 5. Parse and format the script
|
121 |
+
lines = script.strip().split('\n')
|
122 |
+
formatted_script_lines = []
|
123 |
+
for line in lines:
|
124 |
+
line = line.strip()
|
125 |
+
if not line:
|
126 |
+
continue
|
127 |
+
|
128 |
+
# Check if line already has speaker format (e.g., "Speaker 1: ...")
|
129 |
+
if line.startswith('Speaker ') and ':' in line:
|
130 |
+
formatted_script_lines.append(line)
|
131 |
+
else:
|
132 |
+
# Auto-assign speakers in rotation
|
133 |
+
speaker_id = len(formatted_script_lines) % num_speakers
|
134 |
+
formatted_script_lines.append(f"Speaker {speaker_id}: {line}")
|
135 |
+
|
136 |
+
formatted_script = '\n'.join(formatted_script_lines)
|
137 |
+
log += f"π Formatted script with {len(formatted_script_lines)} turns\n"
|
138 |
+
log += "π Processing with VibeVoice...\n"
|
139 |
|
140 |
+
# 6. Prepare inputs for the model
|
141 |
+
# Assuming self.processor is an object available in your class
|
142 |
+
inputs = self.processor(
|
143 |
+
text=[formatted_script],
|
144 |
+
voice_samples=[voice_samples],
|
145 |
+
padding=True,
|
146 |
+
return_tensors="pt",
|
147 |
+
return_attention_mask=True,
|
148 |
+
)
|
149 |
|
150 |
+
# 7. Generate audio
|
151 |
+
start_time = time.time()
|
152 |
+
# Assuming self.model is an object available in your class
|
153 |
+
outputs = self.model.generate(
|
154 |
+
**inputs,
|
155 |
+
max_new_tokens=None,
|
156 |
+
cfg_scale=cfg_scale,
|
157 |
+
tokenizer=self.processor.tokenizer,
|
158 |
+
generation_config={'do_sample': False},
|
159 |
+
verbose=False, # Verbose is off for cleaner logs
|
160 |
+
)
|
161 |
+
generation_time = time.time() - start_time
|
162 |
|
163 |
+
# 8. Extract audio output
|
164 |
+
# The generated audio is often in speech_outputs or a similar attribute
|
165 |
+
if hasattr(outputs, 'speech_outputs') and outputs.speech_outputs[0] is not None:
|
166 |
+
audio_tensor = outputs.speech_outputs[0]
|
167 |
+
audio = audio_tensor.cpu().numpy()
|
168 |
+
else:
|
169 |
+
raise gr.Error("β Error: No audio was generated by the model. Please try again.")
|
|
|
|
|
170 |
|
171 |
+
# Ensure audio is a 1D array
|
172 |
+
if audio.ndim > 1:
|
173 |
+
audio = audio.squeeze()
|
174 |
+
|
175 |
+
sample_rate = 24000 # Standard sample rate for this model
|
176 |
|
177 |
+
# 9. Save the audio file
|
178 |
+
output_dir = "outputs"
|
179 |
+
os.makedirs(output_dir, exist_ok=True)
|
180 |
+
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
|
181 |
+
file_path = os.path.join(output_dir, f"podcast_{timestamp}.wav")
|
182 |
+
|
183 |
+
# Write the NumPy array to a WAV file
|
184 |
+
sf.write(file_path, audio, sample_rate)
|
185 |
+
print(f"πΎ Podcast saved to {file_path}")
|
186 |
|
187 |
+
# 10. Finalize log and return
|
188 |
+
total_duration = len(audio) / sample_rate
|
189 |
+
log += f"β±οΈ Generation completed in {generation_time:.2f} seconds\n"
|
190 |
+
log += f"π΅ Final audio duration: {total_duration:.2f} seconds\n"
|
191 |
+
log += f"β
Successfully saved podcast to: {file_path}\n"
|
192 |
+
|
193 |
+
self.is_generating = False
|
194 |
+
return (sample_rate, audio), log
|
195 |
+
|
196 |
+
except gr.Error as e:
|
197 |
+
# Handle Gradio-specific errors (for user feedback)
|
198 |
+
self.is_generating = False
|
199 |
+
error_msg = f"β Input Error: {str(e)}"
|
200 |
+
print(error_msg)
|
201 |
+
# In Gradio, you would typically return an update to the UI
|
202 |
+
# For a pure function, we re-raise or handle it as needed.
|
203 |
+
# This return signature matches the success case but with error info.
|
204 |
+
return None, error_msg
|
205 |
|
206 |
+
except Exception as e:
|
207 |
+
# Handle all other unexpected errors
|
208 |
+
self.is_generating = False
|
209 |
+
error_msg = f"β An unexpected error occurred: {str(e)}"
|
210 |
+
print(error_msg)
|
211 |
+
import traceback
|
212 |
+
traceback.print_exc()
|
213 |
+
return None, error_msg
|
214 |
|
215 |
|
216 |
|