Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -16,86 +16,174 @@ import spaces
|
|
16 |
from TTS.api import TTS
|
17 |
from TTS.utils.synthesizer import Synthesizer
|
18 |
|
19 |
-
#
|
|
|
|
|
20 |
load_dotenv()
|
21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
|
23 |
# ---------------------------------------------------------------------
|
24 |
# Script Generation Function
|
25 |
# ---------------------------------------------------------------------
|
26 |
@spaces.GPU(duration=100)
|
27 |
def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
|
|
|
|
|
|
|
|
|
28 |
try:
|
29 |
-
|
30 |
-
model = AutoModelForCausalLM.from_pretrained(
|
31 |
-
model_id,
|
32 |
-
use_auth_token=token,
|
33 |
-
torch_dtype=torch.float16,
|
34 |
-
device_map="auto",
|
35 |
-
trust_remote_code=True,
|
36 |
-
)
|
37 |
-
llama_pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)
|
38 |
|
39 |
# System prompt with clear structure instructions
|
40 |
system_prompt = (
|
41 |
-
|
42 |
f"Based on the user's concept and the selected duration of {duration} seconds, produce the following: "
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
)
|
47 |
|
48 |
combined_prompt = f"{system_prompt}\nUser concept: {user_prompt}\nOutput:"
|
49 |
-
result = llama_pipeline(combined_prompt, max_new_tokens=300, do_sample=True, temperature=0.8)
|
50 |
|
51 |
-
#
|
52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
|
59 |
return voice_script, sound_design, music_suggestions
|
|
|
60 |
except Exception as e:
|
61 |
return f"Error generating script: {e}", "", ""
|
62 |
|
|
|
63 |
# ---------------------------------------------------------------------
|
64 |
# Voice-Over Generation Function (Inactive)
|
65 |
# ---------------------------------------------------------------------
|
66 |
@spaces.GPU(duration=100)
|
67 |
def generate_voice(script: str, speaker: str = "default"):
|
|
|
|
|
|
|
68 |
try:
|
69 |
-
# Placeholder for inactive state
|
70 |
return "Voice-over generation is currently inactive."
|
71 |
except Exception as e:
|
72 |
return f"Error: {e}"
|
73 |
|
|
|
74 |
# ---------------------------------------------------------------------
|
75 |
-
# Music Generation Function
|
76 |
# ---------------------------------------------------------------------
|
77 |
@spaces.GPU(duration=100)
|
78 |
def generate_music(prompt: str, audio_length: int):
|
|
|
|
|
|
|
|
|
79 |
try:
|
80 |
-
|
81 |
-
musicgen_model =
|
82 |
-
musicgen_processor = AutoProcessor.from_pretrained("facebook/musicgen-medium")
|
83 |
|
84 |
-
# Move the model to the appropriate device (CUDA or CPU)
|
85 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
86 |
-
|
87 |
-
|
88 |
-
# Prepare inputs
|
89 |
inputs = musicgen_processor(text=[prompt], padding=True, return_tensors="pt").to(device)
|
90 |
|
91 |
-
# Generate music
|
92 |
-
|
|
|
93 |
|
94 |
-
# Process audio data
|
95 |
audio_data = outputs[0, 0].cpu().numpy()
|
|
|
96 |
normalized_audio = (audio_data / max(abs(audio_data)) * 32767).astype("int16")
|
97 |
|
98 |
-
# Save generated music to a file
|
99 |
output_path = f"{tempfile.gettempdir()}/musicgen_medium_generated_music.wav"
|
100 |
write(output_path, 44100, normalized_audio)
|
101 |
|
@@ -106,11 +194,13 @@ def generate_music(prompt: str, audio_length: int):
|
|
106 |
|
107 |
|
108 |
# ---------------------------------------------------------------------
|
109 |
-
# Audio Blending Function
|
110 |
# ---------------------------------------------------------------------
|
111 |
def blend_audio(voice_path: str, music_path: str, ducking: bool):
|
|
|
|
|
|
|
112 |
try:
|
113 |
-
# Placeholder for inactive state
|
114 |
return "Audio blending functionality is currently inactive."
|
115 |
except Exception as e:
|
116 |
return f"Error: {e}"
|
@@ -121,73 +211,92 @@ def blend_audio(voice_path: str, music_path: str, ducking: bool):
|
|
121 |
# ---------------------------------------------------------------------
|
122 |
with gr.Blocks() as demo:
|
123 |
gr.Markdown("""
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
""")
|
128 |
|
129 |
with gr.Tabs():
|
130 |
# Step 1: Generate Script
|
131 |
with gr.Tab("Step 1: Generate Script"):
|
132 |
with gr.Row():
|
133 |
-
user_prompt = gr.Textbox(
|
134 |
-
|
135 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
|
137 |
generate_script_button = gr.Button("Generate Script")
|
138 |
-
script_output = gr.Textbox(label="Generated Voice-Over Script", lines=5)
|
139 |
-
sound_design_output = gr.Textbox(label="Sound Design Suggestions", lines=3)
|
140 |
-
music_suggestion_output = gr.Textbox(label="Music Suggestions", lines=3)
|
141 |
|
142 |
generate_script_button.click(
|
143 |
-
fn=lambda user_prompt, model_id,
|
144 |
inputs=[user_prompt, llama_model_id, duration],
|
145 |
outputs=[script_output, sound_design_output, music_suggestion_output],
|
146 |
)
|
147 |
|
148 |
-
# Step 2: Generate Voice
|
149 |
with gr.Tab("Step 2: Generate Voice"):
|
150 |
gr.Markdown("""
|
151 |
-
|
152 |
-
|
153 |
""")
|
154 |
-
|
155 |
|
156 |
# Step 3: Generate Music
|
157 |
with gr.Tab("Step 3: Generate Music"):
|
158 |
with gr.Row():
|
159 |
-
audio_length = gr.Slider(
|
160 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
161 |
generate_music_button = gr.Button("Generate Music")
|
162 |
-
music_output = gr.Audio(label="Generated Music", type="filepath")
|
163 |
|
164 |
generate_music_button.click(
|
165 |
-
fn=lambda music_suggestion,
|
166 |
inputs=[music_suggestion_output, audio_length],
|
167 |
outputs=[music_output],
|
168 |
)
|
169 |
|
170 |
-
# Step 4: Blend Audio
|
171 |
with gr.Tab("Step 4: Blend Audio"):
|
172 |
gr.Markdown("""
|
173 |
-
|
174 |
-
|
175 |
""")
|
176 |
-
|
177 |
|
|
|
178 |
gr.Markdown("""
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
""")
|
184 |
|
185 |
-
#
|
186 |
gr.HTML("""
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
""")
|
191 |
|
|
|
192 |
demo.launch(debug=True)
|
193 |
-
|
|
|
16 |
from TTS.api import TTS
|
17 |
from TTS.utils.synthesizer import Synthesizer
|
18 |
|
19 |
+
# ---------------------------------------------------------------------
|
20 |
+
# Load Environment Variables
|
21 |
+
# ---------------------------------------------------------------------
|
22 |
load_dotenv()
|
23 |
+
HF_TOKEN = os.getenv("HF_TOKEN")
|
24 |
+
|
25 |
+
# ---------------------------------------------------------------------
|
26 |
+
# Global Model Caches
|
27 |
+
# ---------------------------------------------------------------------
|
28 |
+
# We store models/pipelines in global variables for reuse,
|
29 |
+
# so they are only loaded once.
|
30 |
+
LLAMA_PIPELINES = {}
|
31 |
+
MUSICGEN_MODELS = {}
|
32 |
+
|
33 |
+
# ---------------------------------------------------------------------
|
34 |
+
# Helper Functions
|
35 |
+
# ---------------------------------------------------------------------
|
36 |
+
def get_llama_pipeline(model_id: str, token: str):
|
37 |
+
"""
|
38 |
+
Returns a cached LLaMA pipeline if available; otherwise, loads it.
|
39 |
+
This significantly reduces loading time for repeated calls.
|
40 |
+
"""
|
41 |
+
if model_id in LLAMA_PIPELINES:
|
42 |
+
return LLAMA_PIPELINES[model_id]
|
43 |
+
|
44 |
+
# Load new pipeline and store in cache
|
45 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id, use_auth_token=token)
|
46 |
+
model = AutoModelForCausalLM.from_pretrained(
|
47 |
+
model_id,
|
48 |
+
use_auth_token=token,
|
49 |
+
torch_dtype=torch.float16,
|
50 |
+
device_map="auto",
|
51 |
+
trust_remote_code=True,
|
52 |
+
)
|
53 |
+
text_pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)
|
54 |
+
LLAMA_PIPELINES[model_id] = text_pipeline
|
55 |
+
return text_pipeline
|
56 |
+
|
57 |
+
|
58 |
+
def get_musicgen_model(model_key: str = "facebook/musicgen-medium"):
|
59 |
+
"""
|
60 |
+
Returns a cached MusicGen model if available; otherwise, loads it.
|
61 |
+
"""
|
62 |
+
if model_key in MUSICGEN_MODELS:
|
63 |
+
return MUSICGEN_MODELS[model_key]
|
64 |
+
|
65 |
+
# Load new MusicGen model and store in cache
|
66 |
+
model = MusicgenForConditionalGeneration.from_pretrained(model_key)
|
67 |
+
processor = AutoProcessor.from_pretrained(model_key)
|
68 |
+
|
69 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
70 |
+
model.to(device)
|
71 |
+
|
72 |
+
MUSICGEN_MODELS[model_key] = (model, processor)
|
73 |
+
return model, processor
|
74 |
+
|
75 |
|
76 |
# ---------------------------------------------------------------------
|
77 |
# Script Generation Function
|
78 |
# ---------------------------------------------------------------------
|
79 |
@spaces.GPU(duration=100)
|
80 |
def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
|
81 |
+
"""
|
82 |
+
Generates a script, sound design suggestions, and music ideas from a user prompt.
|
83 |
+
Returns a tuple of strings: (voice_script, sound_design, music_suggestions).
|
84 |
+
"""
|
85 |
try:
|
86 |
+
text_pipeline = get_llama_pipeline(model_id, token)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
87 |
|
88 |
# System prompt with clear structure instructions
|
89 |
system_prompt = (
|
90 |
+
"You are an expert radio imaging producer specializing in sound design and music. "
|
91 |
f"Based on the user's concept and the selected duration of {duration} seconds, produce the following: "
|
92 |
+
"1. A concise voice-over script. Prefix this section with 'Voice-Over Script:'.\n"
|
93 |
+
"2. Suggestions for sound design. Prefix this section with 'Sound Design Suggestions:'.\n"
|
94 |
+
"3. Music styles or track recommendations. Prefix this section with 'Music Suggestions:'."
|
95 |
)
|
96 |
|
97 |
combined_prompt = f"{system_prompt}\nUser concept: {user_prompt}\nOutput:"
|
|
|
98 |
|
99 |
+
# Use inference mode for efficient forward passes
|
100 |
+
with torch.inference_mode():
|
101 |
+
result = text_pipeline(
|
102 |
+
combined_prompt,
|
103 |
+
max_new_tokens=300,
|
104 |
+
do_sample=True,
|
105 |
+
temperature=0.8
|
106 |
+
)
|
107 |
+
|
108 |
+
# LLaMA pipeline returns a list of dicts with "generated_text"
|
109 |
+
generated_text = result[0]["generated_text"]
|
110 |
+
|
111 |
+
# Basic parsing to isolate everything after "Output:"
|
112 |
+
# (in case the model repeated your system prompt).
|
113 |
+
if "Output:" in generated_text:
|
114 |
+
generated_text = generated_text.split("Output:")[-1].strip()
|
115 |
+
|
116 |
+
# Extract sections based on known prefixes
|
117 |
+
voice_script = "No voice-over script found."
|
118 |
+
sound_design = "No sound design suggestions found."
|
119 |
+
music_suggestions = "No music suggestions found."
|
120 |
+
|
121 |
+
if "Voice-Over Script:" in generated_text:
|
122 |
+
parts = generated_text.split("Voice-Over Script:")
|
123 |
+
if len(parts) > 1:
|
124 |
+
# Everything after "Voice-Over Script:" up until next prefix
|
125 |
+
voice_script_part = parts[1]
|
126 |
+
voice_script = voice_script_part.split("Sound Design Suggestions:")[0].strip() \
|
127 |
+
if "Sound Design Suggestions:" in voice_script_part else voice_script_part.strip()
|
128 |
|
129 |
+
if "Sound Design Suggestions:" in generated_text:
|
130 |
+
parts = generated_text.split("Sound Design Suggestions:")
|
131 |
+
if len(parts) > 1:
|
132 |
+
sound_design_part = parts[1]
|
133 |
+
sound_design = sound_design_part.split("Music Suggestions:")[0].strip() \
|
134 |
+
if "Music Suggestions:" in sound_design_part else sound_design_part.strip()
|
135 |
+
|
136 |
+
if "Music Suggestions:" in generated_text:
|
137 |
+
parts = generated_text.split("Music Suggestions:")
|
138 |
+
if len(parts) > 1:
|
139 |
+
music_suggestions = parts[1].strip()
|
140 |
|
141 |
return voice_script, sound_design, music_suggestions
|
142 |
+
|
143 |
except Exception as e:
|
144 |
return f"Error generating script: {e}", "", ""
|
145 |
|
146 |
+
|
147 |
# ---------------------------------------------------------------------
|
148 |
# Voice-Over Generation Function (Inactive)
|
149 |
# ---------------------------------------------------------------------
|
150 |
@spaces.GPU(duration=100)
|
151 |
def generate_voice(script: str, speaker: str = "default"):
|
152 |
+
"""
|
153 |
+
Placeholder for future voice-over generation functionality.
|
154 |
+
"""
|
155 |
try:
|
|
|
156 |
return "Voice-over generation is currently inactive."
|
157 |
except Exception as e:
|
158 |
return f"Error: {e}"
|
159 |
|
160 |
+
|
161 |
# ---------------------------------------------------------------------
|
162 |
+
# Music Generation Function
|
163 |
# ---------------------------------------------------------------------
|
164 |
@spaces.GPU(duration=100)
|
165 |
def generate_music(prompt: str, audio_length: int):
|
166 |
+
"""
|
167 |
+
Generates music from the 'facebook/musicgen-medium' model based on the prompt.
|
168 |
+
Returns the file path to the generated .wav file.
|
169 |
+
"""
|
170 |
try:
|
171 |
+
model_key = "facebook/musicgen-medium"
|
172 |
+
musicgen_model, musicgen_processor = get_musicgen_model(model_key)
|
|
|
173 |
|
|
|
174 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
175 |
+
# Prepare input
|
|
|
|
|
176 |
inputs = musicgen_processor(text=[prompt], padding=True, return_tensors="pt").to(device)
|
177 |
|
178 |
+
# Generate music within inference mode
|
179 |
+
with torch.inference_mode():
|
180 |
+
outputs = musicgen_model.generate(**inputs, max_new_tokens=audio_length)
|
181 |
|
|
|
182 |
audio_data = outputs[0, 0].cpu().numpy()
|
183 |
+
# Normalize audio to int16 format
|
184 |
normalized_audio = (audio_data / max(abs(audio_data)) * 32767).astype("int16")
|
185 |
|
186 |
+
# Save generated music to a temp file
|
187 |
output_path = f"{tempfile.gettempdir()}/musicgen_medium_generated_music.wav"
|
188 |
write(output_path, 44100, normalized_audio)
|
189 |
|
|
|
194 |
|
195 |
|
196 |
# ---------------------------------------------------------------------
|
197 |
+
# Audio Blending Function (Inactive)
|
198 |
# ---------------------------------------------------------------------
|
199 |
def blend_audio(voice_path: str, music_path: str, ducking: bool):
|
200 |
+
"""
|
201 |
+
Placeholder for future audio blending functionality with optional ducking.
|
202 |
+
"""
|
203 |
try:
|
|
|
204 |
return "Audio blending functionality is currently inactive."
|
205 |
except Exception as e:
|
206 |
return f"Error: {e}"
|
|
|
211 |
# ---------------------------------------------------------------------
|
212 |
with gr.Blocks() as demo:
|
213 |
gr.Markdown("""
|
214 |
+
# 🎧 AI Promo Studio 🚀
|
215 |
+
Welcome to **AI Promo Studio**, your one-stop solution for creating stunning and professional radio promos with ease!
|
216 |
+
Whether you're a sound designer, radio producer, or content creator, our AI-driven tools, powered by advanced LLM Llama models, empower you to bring your vision to life in just a few steps.
|
217 |
""")
|
218 |
|
219 |
with gr.Tabs():
|
220 |
# Step 1: Generate Script
|
221 |
with gr.Tab("Step 1: Generate Script"):
|
222 |
with gr.Row():
|
223 |
+
user_prompt = gr.Textbox(
|
224 |
+
label="Promo Idea",
|
225 |
+
placeholder="E.g., A 30-second promo for a morning show...",
|
226 |
+
lines=2
|
227 |
+
)
|
228 |
+
llama_model_id = gr.Textbox(
|
229 |
+
label="LLaMA Model ID",
|
230 |
+
value="meta-llama/Meta-Llama-3-8B-Instruct",
|
231 |
+
placeholder="Enter a valid Hugging Face model ID"
|
232 |
+
)
|
233 |
+
duration = gr.Slider(
|
234 |
+
label="Desired Promo Duration (seconds)",
|
235 |
+
minimum=15,
|
236 |
+
maximum=60,
|
237 |
+
step=15,
|
238 |
+
value=30
|
239 |
+
)
|
240 |
|
241 |
generate_script_button = gr.Button("Generate Script")
|
242 |
+
script_output = gr.Textbox(label="Generated Voice-Over Script", lines=5, interactive=False)
|
243 |
+
sound_design_output = gr.Textbox(label="Sound Design Suggestions", lines=3, interactive=False)
|
244 |
+
music_suggestion_output = gr.Textbox(label="Music Suggestions", lines=3, interactive=False)
|
245 |
|
246 |
generate_script_button.click(
|
247 |
+
fn=lambda user_prompt, model_id, dur: generate_script(user_prompt, model_id, HF_TOKEN, dur),
|
248 |
inputs=[user_prompt, llama_model_id, duration],
|
249 |
outputs=[script_output, sound_design_output, music_suggestion_output],
|
250 |
)
|
251 |
|
252 |
+
# Step 2: Generate Voice (Inactive)
|
253 |
with gr.Tab("Step 2: Generate Voice"):
|
254 |
gr.Markdown("""
|
255 |
+
**Note:** Voice-over generation is currently inactive.
|
256 |
+
This feature will be available in future updates!
|
257 |
""")
|
|
|
258 |
|
259 |
# Step 3: Generate Music
|
260 |
with gr.Tab("Step 3: Generate Music"):
|
261 |
with gr.Row():
|
262 |
+
audio_length = gr.Slider(
|
263 |
+
label="Music Length (tokens)",
|
264 |
+
minimum=128,
|
265 |
+
maximum=1024,
|
266 |
+
step=64,
|
267 |
+
value=512,
|
268 |
+
info="Increase tokens for longer audio, but be mindful of inference time."
|
269 |
+
)
|
270 |
generate_music_button = gr.Button("Generate Music")
|
271 |
+
music_output = gr.Audio(label="Generated Music (WAV)", type="filepath")
|
272 |
|
273 |
generate_music_button.click(
|
274 |
+
fn=lambda music_suggestion, length: generate_music(music_suggestion, length),
|
275 |
inputs=[music_suggestion_output, audio_length],
|
276 |
outputs=[music_output],
|
277 |
)
|
278 |
|
279 |
+
# Step 4: Blend Audio (Inactive)
|
280 |
with gr.Tab("Step 4: Blend Audio"):
|
281 |
gr.Markdown("""
|
282 |
+
**Note:** Audio blending functionality is currently inactive.
|
283 |
+
This feature will be available in future updates!
|
284 |
""")
|
|
|
285 |
|
286 |
+
# Footer / Credits
|
287 |
gr.Markdown("""
|
288 |
+
<hr>
|
289 |
+
<p style="text-align: center; font-size: 0.9em;">
|
290 |
+
Created with ❤️ by <a href="https://bilsimaging.com" target="_blank">bilsimaging.com</a>
|
291 |
+
</p>
|
292 |
""")
|
293 |
|
294 |
+
# Visitor Badge
|
295 |
gr.HTML("""
|
296 |
+
<a href="https://visitorbadge.io/status?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2FBils%2Fradiogold">
|
297 |
+
<img src="https://api.visitorbadge.io/api/visitors?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2FBils%2Fradiogold&countColor=%23263759" />
|
298 |
+
</a>
|
299 |
""")
|
300 |
|
301 |
+
# Launch the Gradio app
|
302 |
demo.launch(debug=True)
|
|