Spaces:
Sleeping
Sleeping
File size: 13,203 Bytes
ce2ed27 232ff56 ce2ed27 232ff56 ce2ed27 232ff56 ce2ed27 232ff56 ce2ed27 232ff56 ce2ed27 232ff56 ce2ed27 232ff56 ce2ed27 232ff56 ce2ed27 232ff56 ce2ed27 232ff56 ce2ed27 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 |
import gradio as gr
from audio_transform_demo import AudioRedTeamOrchestrator
import os
from typing import Optional
# Define the available options for each hyperparameter
TARGET_MODELS = [
"gpt-4o-audio-preview",
"gemini-2.5-pro-preview-05-06",
"gemini-2.0-flash",
"gemini-2.5-flash-preview-04-17",
]
TTS_PROVIDERS = ["kokoro", "smallestai"]
TTS_MODELS = [
"Kokoro-82M",
"lightning",
"lightning-large",
]
VOICE_IDS = [
"af_heart (American, F, Kokoro-82M)",
"karen (British, F, lightning)",
"rebecca (American, F, lightning)",
"chetan (Indian, M, lightning)",
"george (American, M, lightning)",
"solomon (British, M, lightning-large)",
"saina (Indian, F, lightning)",
"angela (British, F, lightning-large)",
"nyah (Australian, F, lightning-large)",
]
TRANSFORM_TYPES = ["none", "speed", "pitch", "echo", "reverb", "noise"]
def create_transform_ui():
"""Create dynamic UI components for transform parameters"""
with gr.Blocks() as transform_block:
transform_type = gr.Dropdown(
choices=TRANSFORM_TYPES,
label="Transform Type",
value="none",
info="Select the type of audio transformation to apply",
)
# Speed transform parameters
speed_rate = gr.Slider(
minimum=0.5,
maximum=2.0,
value=1.0,
step=0.1,
label="Speed Rate",
info="Rate of the speed effect (1.0 = original speed)",
visible=False,
)
# Pitch transform parameters
pitch_steps = gr.Slider(
minimum=-12,
maximum=12,
value=4,
step=1,
label="Pitch Steps",
info="Steps to shift the pitch (in semitones)",
visible=False,
)
pitch_bins = gr.Slider(
minimum=6,
maximum=24,
value=12,
step=1,
label="Bins per Octave",
info="Steps per octave (12 = standard semitones)",
visible=False,
)
# Echo transform parameters
echo_delay = gr.Slider(
minimum=50,
maximum=1000,
value=250,
step=1,
label="Echo Delay (ms)",
info="Delay in milliseconds for the echo effect",
visible=False,
)
echo_volume = gr.Slider(
minimum=-20,
maximum=0,
value=-6,
step=1,
label="Echo Volume (dB)",
info="Volume change in dB for the echo effect",
visible=False,
)
# Reverb transform parameters
reverb_rir = gr.File(
label="Room Impulse Response File",
file_types=[".wav"],
visible=False,
)
reverb_method = gr.Radio(
choices=["fft", "conv1d"],
value="fft",
label="Convolution Method",
info="Method to use for applying reverb",
visible=False,
)
# Noise transform parameters
noise_file = gr.File(
label="Noise File",
file_types=[".wav"],
visible=False,
)
noise_volume = gr.Slider(
minimum=-20,
maximum=40,
value=-10,
step=1,
label="Noise Volume (dB)",
info="Volume change in dB for the noise effect",
visible=False,
)
# Function to update visible components based on transform type
def update_transform_ui(transform_type):
return [
speed_rate.update(visible=transform_type == "speed"),
pitch_steps.update(visible=transform_type == "pitch"),
pitch_bins.update(visible=transform_type == "pitch"),
echo_delay.update(visible=transform_type == "echo"),
echo_volume.update(visible=transform_type == "echo"),
reverb_rir.update(visible=transform_type == "reverb"),
reverb_method.update(visible=transform_type == "reverb"),
noise_file.update(visible=transform_type == "noise"),
noise_volume.update(visible=transform_type == "noise"),
]
# Connect the transform type dropdown to update the UI
transform_type.change(
fn=update_transform_ui,
inputs=[transform_type],
outputs=[
speed_rate,
pitch_steps,
pitch_bins,
echo_delay,
echo_volume,
reverb_rir,
reverb_method,
noise_file,
noise_volume,
],
)
return transform_block, {
"speed": {"rate": speed_rate},
"pitch": {"n_steps": pitch_steps, "bins_per_octave": pitch_bins},
"echo": {"delay": echo_delay, "volume": echo_volume},
"reverb": {"rir_path": reverb_rir, "conv_method": reverb_method},
"noise": {"noise_path": noise_file, "volume": noise_volume},
}
def create_redteam_demo():
def process_attack(
prompt,
target_model,
tts_provider,
tts_model,
voice_id,
transform_type,
speed_rate: Optional[float] = None,
pitch_steps: Optional[int] = None,
pitch_bins: Optional[int] = None,
echo_delay: Optional[int] = None,
echo_volume: Optional[int] = None,
reverb_rir: Optional[str] = None,
reverb_method: Optional[str] = None,
noise_file: Optional[str] = None,
noise_volume: Optional[int] = None,
):
# Prepare transform parameters
# print("AAAA")
transform_kwargs = None
if transform_type != "none":
if transform_type == "speed":
transform_kwargs = {"rate": speed_rate}
elif transform_type == "pitch":
transform_kwargs = {
"n_steps": pitch_steps,
"bins_per_octave": pitch_bins,
}
elif transform_type == "echo":
transform_kwargs = {"delay": echo_delay, "volume": echo_volume}
elif transform_type == "reverb":
transform_kwargs = {
"rir_path": reverb_rir.name if reverb_rir else None,
"conv_method": reverb_method,
}
elif transform_type == "noise":
transform_kwargs = {
"noise_path": noise_file.name if noise_file else None,
"volume": noise_volume,
}
# print("BBBB")
voice_id = voice_id.split("(")[0].strip()
print("Voice ID: ", voice_id)
# Initialize the orchestrator with selected parameters
orchestrator = AudioRedTeamOrchestrator(
tts_provider=tts_provider,
model_name=target_model,
voice_id=voice_id,
tts_model=tts_model,
)
# print("CCCC")
# Create a temporary directory for saving files
save_dir = "temp_audio_files"
os.makedirs(save_dir, exist_ok=True)
# Generate a unique ID for this attack
prompt_id = "attack_1"
# print("DDDD")
# Run the attack
result = orchestrator.attack(
prompt=prompt,
prompt_id=prompt_id,
save_dir=save_dir,
model_name=target_model,
generate_audio=False,
transform_type=transform_type if transform_type != "none" else None,
transform_kwargs=transform_kwargs,
)
# print("EEEE")
print("Attack run successfully")
if result:
audio_path = f"{save_dir}/{prompt_id}.wav"
if os.path.exists(audio_path):
print("Audio file exists")
return result, audio_path
else:
return result, None
else:
return {"Evaluation Result": "Attack failed or no response generated"}, None
with gr.Blocks() as demo:
prompt = gr.Textbox(label="Attack Prompt", lines=3)
target_model = gr.Dropdown(
choices=TARGET_MODELS, label="Target Model", value=TARGET_MODELS[0]
)
tts_provider = gr.Dropdown(
choices=TTS_PROVIDERS, label="TTS Provider", value=TTS_PROVIDERS[0]
)
tts_model = gr.Dropdown(
choices=TTS_MODELS, label="TTS Model", value=TTS_MODELS[0]
)
voice_id = gr.Dropdown(choices=VOICE_IDS, label="Voice ID", value=VOICE_IDS[0])
transform_type = gr.Dropdown(
choices=TRANSFORM_TYPES,
label="Transform Type",
value="none",
interactive=True,
info="Select the type of audio transformation to apply",
)
speed_rate = gr.Slider(
minimum=0.5,
maximum=2.0,
value=1.0,
step=0.1,
label="Speed Rate",
interactive=True,
info="Rate of the speed effect (1.0 = original speed)",
visible=False,
)
pitch_steps = gr.Slider(
minimum=-12,
maximum=12,
value=4,
step=1,
label="Pitch Steps",
interactive=True,
info="Steps to shift the pitch (in semitones)",
visible=False,
)
pitch_bins = gr.Slider(
minimum=6,
maximum=24,
value=12,
step=1,
label="Bins per Octave",
interactive=True,
info="Steps per octave (12 = standard semitones)",
visible=False,
)
echo_delay = gr.Slider(
minimum=50,
maximum=1000,
value=250,
step=1,
label="Echo Delay (ms)",
interactive=True,
info="Delay in milliseconds for the echo effect",
visible=False,
)
echo_volume = gr.Slider(
minimum=-20,
maximum=0,
value=-6,
step=1,
label="Echo Volume (dB)",
interactive=True,
info="Volume change in dB for the echo effect",
visible=False,
)
reverb_rir = gr.File(
label="Room Impulse Response File",
file_types=[".wav"],
interactive=True,
visible=False,
)
reverb_method = gr.Radio(
choices=["fft", "conv1d"],
value="fft",
label="Convolution Method",
interactive=True,
info="Method to use for applying reverb",
visible=False,
)
noise_file = gr.File(
label="Noise File",
file_types=[".wav"],
interactive=True,
visible=False,
)
noise_volume = gr.Slider(
minimum=-20,
maximum=40,
value=-10,
step=1,
label="Noise Volume (dB)",
interactive=True,
info="Volume change in dB for the noise effect",
visible=False,
)
# Function to update visible components based on transform type
def update_transform_ui(transform_type):
return [
gr.update(visible=transform_type == "speed"),
gr.update(visible=transform_type == "pitch"),
gr.update(visible=transform_type == "pitch"),
gr.update(visible=transform_type == "echo"),
gr.update(visible=transform_type == "echo"),
gr.update(visible=transform_type == "reverb"),
gr.update(visible=transform_type == "reverb"),
gr.update(visible=transform_type == "noise"),
gr.update(visible=transform_type == "noise"),
]
# print("FFFF")
transform_type.change(
fn=update_transform_ui,
inputs=[transform_type],
outputs=[
speed_rate,
pitch_steps,
pitch_bins,
echo_delay,
echo_volume,
reverb_rir,
reverb_method,
noise_file,
noise_volume,
],
)
# print("GGGG")
btn = gr.Button("Run Attack")
out_json = gr.JSON()
out_audio = gr.Audio()
# print("HHHH")
btn.click(
fn=process_attack,
inputs=[
prompt,
target_model,
tts_provider,
tts_model,
voice_id,
transform_type,
speed_rate,
pitch_steps,
pitch_bins,
echo_delay,
echo_volume,
reverb_rir,
reverb_method,
noise_file,
noise_volume,
],
outputs=[out_json, out_audio],
)
return demo
if __name__ == "__main__":
demo = create_redteam_demo()
demo.launch(share=True)
|