Spaces:
Sleeping
Sleeping
Commit
·
640dd0e
1
Parent(s):
af81629
Code changes
Browse files
app.py
CHANGED
|
@@ -1,96 +1,149 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
import numpy as np
|
| 3 |
-
import
|
| 4 |
-
import torchaudio
|
| 5 |
-
import threading
|
| 6 |
import queue
|
|
|
|
| 7 |
import time
|
|
|
|
| 8 |
import os
|
| 9 |
import urllib.request
|
|
|
|
| 10 |
from scipy.spatial.distance import cosine
|
| 11 |
-
from
|
| 12 |
-
import
|
| 13 |
-
import librosa
|
| 14 |
|
| 15 |
-
#
|
| 16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
TRANSCRIPTION_LANGUAGE = "en"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
DEFAULT_CHANGE_THRESHOLD = 0.7
|
| 19 |
EMBEDDING_HISTORY_SIZE = 5
|
| 20 |
MIN_SEGMENT_DURATION = 1.0
|
| 21 |
DEFAULT_MAX_SPEAKERS = 4
|
| 22 |
-
ABSOLUTE_MAX_SPEAKERS =
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
SAMPLE_RATE = 16000
|
|
|
|
|
|
|
| 24 |
|
| 25 |
-
# Speaker colors
|
| 26 |
SPEAKER_COLORS = [
|
| 27 |
-
"#
|
| 28 |
-
"#
|
| 29 |
-
"#
|
| 30 |
-
"#
|
| 31 |
-
"#
|
| 32 |
-
"#
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
]
|
| 34 |
|
| 35 |
SPEAKER_COLOR_NAMES = [
|
| 36 |
-
"
|
|
|
|
| 37 |
]
|
| 38 |
|
| 39 |
|
| 40 |
class SpeechBrainEncoder:
|
| 41 |
-
"""
|
| 42 |
def __init__(self, device="cpu"):
|
| 43 |
self.device = device
|
| 44 |
-
self.
|
| 45 |
-
self.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
|
|
|
|
|
|
|
| 47 |
def load_model(self):
|
| 48 |
-
"""
|
| 49 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
|
| 51 |
def embed_utterance(self, audio, sr=16000):
|
| 52 |
-
"""Extract
|
|
|
|
|
|
|
|
|
|
| 53 |
try:
|
| 54 |
if isinstance(audio, np.ndarray):
|
| 55 |
-
waveform = torch.tensor(audio, dtype=torch.float32)
|
| 56 |
else:
|
| 57 |
-
waveform = audio
|
| 58 |
-
|
| 59 |
-
if len(waveform.shape) == 1:
|
| 60 |
-
waveform = waveform.unsqueeze(0)
|
| 61 |
|
| 62 |
-
# Resample if needed
|
| 63 |
if sr != 16000:
|
| 64 |
waveform = torchaudio.functional.resample(waveform, orig_freq=sr, new_freq=16000)
|
| 65 |
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 72 |
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
embedding = mfcc.mean(dim=2).flatten()
|
| 76 |
|
| 77 |
-
|
| 78 |
-
if len(embedding) > self.embedding_dim:
|
| 79 |
-
embedding = embedding[:self.embedding_dim]
|
| 80 |
-
elif len(embedding) < self.embedding_dim:
|
| 81 |
-
padding = torch.zeros(self.embedding_dim - len(embedding))
|
| 82 |
-
embedding = torch.cat([embedding, padding])
|
| 83 |
-
|
| 84 |
-
return embedding.numpy()
|
| 85 |
|
|
|
|
| 86 |
except Exception as e:
|
| 87 |
-
print(f"
|
| 88 |
-
return np.
|
| 89 |
|
| 90 |
|
| 91 |
class SpeakerChangeDetector:
|
| 92 |
-
"""Speaker change detector
|
| 93 |
-
def __init__(self, embedding_dim=
|
| 94 |
self.embedding_dim = embedding_dim
|
| 95 |
self.change_threshold = change_threshold
|
| 96 |
self.max_speakers = min(max_speakers, ABSOLUTE_MAX_SPEAKERS)
|
|
@@ -110,6 +163,7 @@ class SpeakerChangeDetector:
|
|
| 110 |
for speaker_id in list(self.active_speakers):
|
| 111 |
if speaker_id >= new_max:
|
| 112 |
self.active_speakers.discard(speaker_id)
|
|
|
|
| 113 |
if self.current_speaker >= new_max:
|
| 114 |
self.current_speaker = 0
|
| 115 |
|
|
@@ -161,6 +215,7 @@ class SpeakerChangeDetector:
|
|
| 161 |
|
| 162 |
if speaker_mean is not None:
|
| 163 |
speaker_similarity = 1.0 - cosine(embedding, speaker_mean)
|
|
|
|
| 164 |
if speaker_similarity > best_similarity:
|
| 165 |
best_similarity = speaker_similarity
|
| 166 |
best_speaker = speaker_id
|
|
@@ -201,317 +256,481 @@ class SpeakerChangeDetector:
|
|
| 201 |
if 0 <= speaker_id < len(SPEAKER_COLORS):
|
| 202 |
return SPEAKER_COLORS[speaker_id]
|
| 203 |
return "#FFFFFF"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 204 |
|
| 205 |
|
| 206 |
-
class
|
| 207 |
-
"""Main class for real-time ASR with speaker diarization"""
|
| 208 |
def __init__(self):
|
| 209 |
-
self.encoder =
|
| 210 |
-
self.
|
| 211 |
-
self.speaker_detector =
|
| 212 |
-
self.
|
| 213 |
-
self.
|
| 214 |
-
self.
|
| 215 |
-
|
| 216 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 217 |
try:
|
| 218 |
-
|
| 219 |
-
|
| 220 |
-
|
| 221 |
-
|
| 222 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 223 |
|
| 224 |
-
def
|
| 225 |
-
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 226 |
try:
|
| 227 |
-
|
| 228 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 229 |
|
| 230 |
-
#
|
| 231 |
-
|
| 232 |
-
|
|
|
|
| 233 |
|
| 234 |
-
|
| 235 |
-
|
|
|
|
| 236 |
|
| 237 |
-
#
|
| 238 |
-
|
| 239 |
-
|
| 240 |
-
audio_data = audio_data / np.abs(audio_data).max()
|
| 241 |
|
| 242 |
-
|
| 243 |
-
if sr != 16000:
|
| 244 |
-
audio_data = librosa.resample(audio_data, orig_sr=sr, target_sr=16000)
|
| 245 |
|
| 246 |
-
|
| 247 |
-
|
| 248 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 249 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 250 |
except Exception as e:
|
| 251 |
print(f"Transcription error: {e}")
|
| 252 |
-
return ""
|
| 253 |
|
| 254 |
-
def
|
| 255 |
-
"""
|
| 256 |
-
|
|
|
|
|
|
|
|
|
|
| 257 |
|
| 258 |
-
def
|
| 259 |
-
"""
|
| 260 |
-
|
| 261 |
-
|
| 262 |
-
|
| 263 |
-
|
| 264 |
-
|
| 265 |
-
|
| 266 |
-
if
|
| 267 |
-
|
| 268 |
-
|
| 269 |
-
|
| 270 |
-
|
| 271 |
-
|
| 272 |
-
# Detect speaker
|
| 273 |
-
speaker_id, similarity = self.speaker_detector.add_embedding(embedding)
|
| 274 |
|
| 275 |
-
return
|
| 276 |
|
| 277 |
-
def
|
| 278 |
-
"""Update
|
| 279 |
-
|
| 280 |
-
|
| 281 |
-
|
| 282 |
-
entry = {
|
| 283 |
-
"speaker": speaker_name,
|
| 284 |
-
"text": transcription,
|
| 285 |
-
"color": color,
|
| 286 |
-
"timestamp": time.time()
|
| 287 |
-
}
|
| 288 |
|
| 289 |
-
self.
|
| 290 |
-
|
| 291 |
-
|
| 292 |
-
def format_conversation_html(self):
|
| 293 |
-
"""Format conversation history as HTML"""
|
| 294 |
-
if not self.conversation_history:
|
| 295 |
-
return "<p><i>No conversation yet. Start speaking to see real-time transcription with speaker diarization.</i></p>"
|
| 296 |
-
|
| 297 |
-
html_parts = []
|
| 298 |
-
for entry in self.conversation_history:
|
| 299 |
-
html_parts.append(
|
| 300 |
-
f'<p><span style="color: {entry["color"]}; font-weight: bold;">'
|
| 301 |
-
f'{entry["speaker"]}:</span> {entry["text"]}</p>'
|
| 302 |
-
)
|
| 303 |
|
| 304 |
-
return ""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 305 |
|
| 306 |
def get_status_info(self):
|
| 307 |
"""Get current status information"""
|
| 308 |
-
|
| 309 |
-
"
|
| 310 |
-
|
| 311 |
-
|
| 312 |
-
|
| 313 |
-
|
| 314 |
-
|
| 315 |
-
|
| 316 |
-
|
| 317 |
-
|
| 318 |
-
|
| 319 |
-
|
| 320 |
-
|
| 321 |
-
|
| 322 |
-
|
| 323 |
-
|
| 324 |
-
|
| 325 |
-
|
| 326 |
-
|
| 327 |
-
|
| 328 |
-
|
|
|
|
|
|
|
|
|
|
| 329 |
|
| 330 |
|
| 331 |
# Global instance
|
| 332 |
-
|
| 333 |
|
| 334 |
|
| 335 |
-
def
|
| 336 |
-
"""
|
| 337 |
-
|
| 338 |
-
|
| 339 |
-
|
| 340 |
-
|
| 341 |
-
|
| 342 |
-
# Update parameters
|
| 343 |
-
asr_system.set_parameters(threshold, max_speakers)
|
| 344 |
-
|
| 345 |
-
try:
|
| 346 |
-
# Process the audio segment
|
| 347 |
-
sr, audio_array = audio_data
|
| 348 |
-
|
| 349 |
-
# Convert to float32 and normalize
|
| 350 |
-
if audio_array.dtype != np.float32:
|
| 351 |
-
audio_array = audio_array.astype(np.float32)
|
| 352 |
-
if audio_array.dtype == np.int16:
|
| 353 |
-
audio_array = audio_array / 32768.0
|
| 354 |
-
elif audio_array.dtype == np.int32:
|
| 355 |
-
audio_array = audio_array / 2147483648.0
|
| 356 |
-
|
| 357 |
-
# Process the audio segment
|
| 358 |
-
transcription, speaker_id, similarity = asr_system.process_audio_segment(audio_array, sr)
|
| 359 |
-
|
| 360 |
-
if transcription and speaker_id is not None:
|
| 361 |
-
# Update conversation
|
| 362 |
-
asr_system.update_conversation(transcription, speaker_id)
|
| 363 |
-
|
| 364 |
-
except Exception as e:
|
| 365 |
-
print(f"Error processing audio: {e}")
|
| 366 |
-
|
| 367 |
-
return asr_system.format_conversation_html(), get_status_display()
|
| 368 |
|
| 369 |
|
| 370 |
-
def
|
| 371 |
-
"""
|
| 372 |
-
|
| 373 |
-
|
| 374 |
-
|
| 375 |
-
|
| 376 |
-
|
| 377 |
-
|
| 378 |
-
Active Speakers: {status['active_speakers']} / {status['max_speakers']}<br>
|
| 379 |
-
Total Segments: {status['total_segments']}<br>
|
| 380 |
-
Threshold: {status['threshold']:.2f}<br>
|
| 381 |
-
</div>
|
| 382 |
-
"""
|
| 383 |
-
|
| 384 |
-
return status_html
|
| 385 |
|
| 386 |
|
| 387 |
def clear_conversation():
|
| 388 |
"""Clear the conversation"""
|
| 389 |
-
|
| 390 |
-
|
| 391 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 392 |
|
|
|
|
|
|
|
|
|
|
| 393 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 394 |
def create_interface():
|
| 395 |
-
""
|
| 396 |
-
|
| 397 |
-
|
| 398 |
-
title="Real-time ASR with Speaker Diarization",
|
| 399 |
-
theme=gr.themes.Soft(),
|
| 400 |
-
css="""
|
| 401 |
-
.conversation-box {
|
| 402 |
-
height: 400px;
|
| 403 |
-
overflow-y: auto;
|
| 404 |
-
border: 1px solid #ddd;
|
| 405 |
-
padding: 10px;
|
| 406 |
-
background-color: #f9f9f9;
|
| 407 |
-
}
|
| 408 |
-
.status-box {
|
| 409 |
-
border: 1px solid #ccc;
|
| 410 |
-
padding: 10px;
|
| 411 |
-
background-color: #f0f0f0;
|
| 412 |
-
}
|
| 413 |
-
"""
|
| 414 |
-
) as demo:
|
| 415 |
-
|
| 416 |
-
gr.Markdown(
|
| 417 |
-
"""
|
| 418 |
-
# 🎤 Real-time ASR with Live Speaker Diarization
|
| 419 |
-
|
| 420 |
-
This application provides real-time speech recognition with speaker diarization.
|
| 421 |
-
It can distinguish between different speakers and display their conversations in different colors.
|
| 422 |
-
|
| 423 |
-
**Instructions:**
|
| 424 |
-
1. Adjust the speaker change threshold and maximum speakers
|
| 425 |
-
2. Click the microphone button to start recording
|
| 426 |
-
3. Speak naturally - the system will detect speaker changes and transcribe speech
|
| 427 |
-
4. Each speaker will be assigned a different color
|
| 428 |
-
"""
|
| 429 |
-
)
|
| 430 |
|
| 431 |
with gr.Row():
|
| 432 |
-
with gr.Column(scale=
|
| 433 |
# Main conversation display
|
| 434 |
-
|
| 435 |
-
value="<
|
| 436 |
-
|
| 437 |
)
|
| 438 |
|
| 439 |
-
#
|
| 440 |
-
|
| 441 |
-
|
| 442 |
-
|
| 443 |
-
|
| 444 |
-
|
| 445 |
-
)
|
| 446 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 447 |
with gr.Column(scale=1):
|
| 448 |
-
#
|
| 449 |
-
gr.Markdown("
|
| 450 |
|
| 451 |
threshold_slider = gr.Slider(
|
| 452 |
minimum=0.1,
|
| 453 |
-
maximum=0.
|
| 454 |
-
value=DEFAULT_CHANGE_THRESHOLD,
|
| 455 |
step=0.05,
|
| 456 |
-
|
| 457 |
-
|
|
|
|
| 458 |
)
|
| 459 |
|
| 460 |
max_speakers_slider = gr.Slider(
|
| 461 |
minimum=2,
|
| 462 |
maximum=ABSOLUTE_MAX_SPEAKERS,
|
| 463 |
-
value=DEFAULT_MAX_SPEAKERS,
|
| 464 |
step=1,
|
| 465 |
-
|
| 466 |
-
|
| 467 |
)
|
| 468 |
|
| 469 |
-
|
| 470 |
-
|
| 471 |
-
# Status display
|
| 472 |
-
gr.Markdown("### Status")
|
| 473 |
-
status_display = gr.HTML(
|
| 474 |
-
value=get_status_display(),
|
| 475 |
-
elem_classes=["status-box"]
|
| 476 |
-
)
|
| 477 |
|
| 478 |
# Speaker color legend
|
| 479 |
-
gr.Markdown("
|
| 480 |
-
|
| 481 |
-
for i in
|
| 482 |
-
|
| 483 |
-
name = SPEAKER_COLOR_NAMES[i]
|
| 484 |
-
legend_html += f'<p><span style="color: {color}; font-weight: bold;">● Speaker {i+1} ({name})</span></p>'
|
| 485 |
|
| 486 |
-
gr.HTML(
|
|
|
|
|
|
|
|
|
|
|
|
|
| 487 |
|
| 488 |
# Event handlers
|
| 489 |
-
|
| 490 |
-
|
| 491 |
-
|
| 492 |
-
|
| 493 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 494 |
)
|
| 495 |
|
| 496 |
clear_btn.click(
|
| 497 |
-
|
| 498 |
-
outputs=[
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 499 |
)
|
| 500 |
|
| 501 |
-
#
|
| 502 |
-
|
| 503 |
-
|
| 504 |
-
|
| 505 |
-
|
| 506 |
)
|
| 507 |
|
| 508 |
-
return
|
| 509 |
|
| 510 |
|
| 511 |
if __name__ == "__main__":
|
| 512 |
-
|
| 513 |
-
|
| 514 |
-
demo.launch(
|
| 515 |
server_name="0.0.0.0",
|
| 516 |
server_port=7860,
|
| 517 |
share=True
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import numpy as np
|
| 3 |
+
import soundcard as sc
|
|
|
|
|
|
|
| 4 |
import queue
|
| 5 |
+
import torch
|
| 6 |
import time
|
| 7 |
+
import threading
|
| 8 |
import os
|
| 9 |
import urllib.request
|
| 10 |
+
import torchaudio
|
| 11 |
from scipy.spatial.distance import cosine
|
| 12 |
+
from RealtimeSTT import AudioToTextRecorder
|
| 13 |
+
import json
|
|
|
|
| 14 |
|
| 15 |
+
# Simplified configuration parameters
|
| 16 |
+
SILENCE_THRESHS = [0, 0.4]
|
| 17 |
+
FINAL_TRANSCRIPTION_MODEL = "distil-large-v3"
|
| 18 |
+
FINAL_BEAM_SIZE = 5
|
| 19 |
+
REALTIME_TRANSCRIPTION_MODEL = "distil-small.en"
|
| 20 |
+
REALTIME_BEAM_SIZE = 5
|
| 21 |
TRANSCRIPTION_LANGUAGE = "en"
|
| 22 |
+
SILERO_SENSITIVITY = 0.4
|
| 23 |
+
WEBRTC_SENSITIVITY = 3
|
| 24 |
+
MIN_LENGTH_OF_RECORDING = 0.7
|
| 25 |
+
PRE_RECORDING_BUFFER_DURATION = 0.35
|
| 26 |
+
|
| 27 |
+
# Speaker change detection parameters
|
| 28 |
DEFAULT_CHANGE_THRESHOLD = 0.7
|
| 29 |
EMBEDDING_HISTORY_SIZE = 5
|
| 30 |
MIN_SEGMENT_DURATION = 1.0
|
| 31 |
DEFAULT_MAX_SPEAKERS = 4
|
| 32 |
+
ABSOLUTE_MAX_SPEAKERS = 10
|
| 33 |
+
|
| 34 |
+
# Global variables
|
| 35 |
+
FAST_SENTENCE_END = True
|
| 36 |
+
USE_MICROPHONE = False
|
| 37 |
SAMPLE_RATE = 16000
|
| 38 |
+
BUFFER_SIZE = 512
|
| 39 |
+
CHANNELS = 1
|
| 40 |
|
| 41 |
+
# Speaker colors
|
| 42 |
SPEAKER_COLORS = [
|
| 43 |
+
"#FFFF00", # Yellow
|
| 44 |
+
"#FF0000", # Red
|
| 45 |
+
"#00FF00", # Green
|
| 46 |
+
"#00FFFF", # Cyan
|
| 47 |
+
"#FF00FF", # Magenta
|
| 48 |
+
"#0000FF", # Blue
|
| 49 |
+
"#FF8000", # Orange
|
| 50 |
+
"#00FF80", # Spring Green
|
| 51 |
+
"#8000FF", # Purple
|
| 52 |
+
"#FFFFFF", # White
|
| 53 |
]
|
| 54 |
|
| 55 |
SPEAKER_COLOR_NAMES = [
|
| 56 |
+
"Yellow", "Red", "Green", "Cyan", "Magenta",
|
| 57 |
+
"Blue", "Orange", "Spring Green", "Purple", "White"
|
| 58 |
]
|
| 59 |
|
| 60 |
|
| 61 |
class SpeechBrainEncoder:
|
| 62 |
+
"""ECAPA-TDNN encoder from SpeechBrain for speaker embeddings"""
|
| 63 |
def __init__(self, device="cpu"):
|
| 64 |
self.device = device
|
| 65 |
+
self.model = None
|
| 66 |
+
self.embedding_dim = 192
|
| 67 |
+
self.model_loaded = False
|
| 68 |
+
self.cache_dir = os.path.join(os.path.expanduser("~"), ".cache", "speechbrain")
|
| 69 |
+
os.makedirs(self.cache_dir, exist_ok=True)
|
| 70 |
+
|
| 71 |
+
def _download_model(self):
|
| 72 |
+
"""Download pre-trained SpeechBrain ECAPA-TDNN model if not present"""
|
| 73 |
+
model_url = "https://huggingface.co/speechbrain/spkrec-ecapa-voxceleb/resolve/main/embedding_model.ckpt"
|
| 74 |
+
model_path = os.path.join(self.cache_dir, "embedding_model.ckpt")
|
| 75 |
+
|
| 76 |
+
if not os.path.exists(model_path):
|
| 77 |
+
print(f"Downloading ECAPA-TDNN model to {model_path}...")
|
| 78 |
+
urllib.request.urlretrieve(model_url, model_path)
|
| 79 |
|
| 80 |
+
return model_path
|
| 81 |
+
|
| 82 |
def load_model(self):
|
| 83 |
+
"""Load the ECAPA-TDNN model"""
|
| 84 |
+
try:
|
| 85 |
+
from speechbrain.pretrained import EncoderClassifier
|
| 86 |
+
|
| 87 |
+
model_path = self._download_model()
|
| 88 |
+
|
| 89 |
+
self.model = EncoderClassifier.from_hparams(
|
| 90 |
+
source="speechbrain/spkrec-ecapa-voxceleb",
|
| 91 |
+
savedir=self.cache_dir,
|
| 92 |
+
run_opts={"device": self.device}
|
| 93 |
+
)
|
| 94 |
+
|
| 95 |
+
self.model_loaded = True
|
| 96 |
+
return True
|
| 97 |
+
except Exception as e:
|
| 98 |
+
print(f"Error loading ECAPA-TDNN model: {e}")
|
| 99 |
+
return False
|
| 100 |
|
| 101 |
def embed_utterance(self, audio, sr=16000):
|
| 102 |
+
"""Extract speaker embedding from audio"""
|
| 103 |
+
if not self.model_loaded:
|
| 104 |
+
raise ValueError("Model not loaded. Call load_model() first.")
|
| 105 |
+
|
| 106 |
try:
|
| 107 |
if isinstance(audio, np.ndarray):
|
| 108 |
+
waveform = torch.tensor(audio, dtype=torch.float32).unsqueeze(0)
|
| 109 |
else:
|
| 110 |
+
waveform = audio.unsqueeze(0)
|
|
|
|
|
|
|
|
|
|
| 111 |
|
|
|
|
| 112 |
if sr != 16000:
|
| 113 |
waveform = torchaudio.functional.resample(waveform, orig_freq=sr, new_freq=16000)
|
| 114 |
|
| 115 |
+
with torch.no_grad():
|
| 116 |
+
embedding = self.model.encode_batch(waveform)
|
| 117 |
+
|
| 118 |
+
return embedding.squeeze().cpu().numpy()
|
| 119 |
+
except Exception as e:
|
| 120 |
+
print(f"Error extracting embedding: {e}")
|
| 121 |
+
return np.zeros(self.embedding_dim)
|
| 122 |
+
|
| 123 |
+
|
| 124 |
+
class AudioProcessor:
|
| 125 |
+
"""Processes audio data to extract speaker embeddings"""
|
| 126 |
+
def __init__(self, encoder):
|
| 127 |
+
self.encoder = encoder
|
| 128 |
+
|
| 129 |
+
def extract_embedding(self, audio_int16):
|
| 130 |
+
try:
|
| 131 |
+
float_audio = audio_int16.astype(np.float32) / 32768.0
|
| 132 |
|
| 133 |
+
if np.abs(float_audio).max() > 1.0:
|
| 134 |
+
float_audio = float_audio / np.abs(float_audio).max()
|
|
|
|
| 135 |
|
| 136 |
+
embedding = self.encoder.embed_utterance(float_audio)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 137 |
|
| 138 |
+
return embedding
|
| 139 |
except Exception as e:
|
| 140 |
+
print(f"Embedding extraction error: {e}")
|
| 141 |
+
return np.zeros(self.encoder.embedding_dim)
|
| 142 |
|
| 143 |
|
| 144 |
class SpeakerChangeDetector:
|
| 145 |
+
"""Speaker change detector that supports a configurable number of speakers"""
|
| 146 |
+
def __init__(self, embedding_dim=192, change_threshold=DEFAULT_CHANGE_THRESHOLD, max_speakers=DEFAULT_MAX_SPEAKERS):
|
| 147 |
self.embedding_dim = embedding_dim
|
| 148 |
self.change_threshold = change_threshold
|
| 149 |
self.max_speakers = min(max_speakers, ABSOLUTE_MAX_SPEAKERS)
|
|
|
|
| 163 |
for speaker_id in list(self.active_speakers):
|
| 164 |
if speaker_id >= new_max:
|
| 165 |
self.active_speakers.discard(speaker_id)
|
| 166 |
+
|
| 167 |
if self.current_speaker >= new_max:
|
| 168 |
self.current_speaker = 0
|
| 169 |
|
|
|
|
| 215 |
|
| 216 |
if speaker_mean is not None:
|
| 217 |
speaker_similarity = 1.0 - cosine(embedding, speaker_mean)
|
| 218 |
+
|
| 219 |
if speaker_similarity > best_similarity:
|
| 220 |
best_similarity = speaker_similarity
|
| 221 |
best_speaker = speaker_id
|
|
|
|
| 256 |
if 0 <= speaker_id < len(SPEAKER_COLORS):
|
| 257 |
return SPEAKER_COLORS[speaker_id]
|
| 258 |
return "#FFFFFF"
|
| 259 |
+
|
| 260 |
+
def get_status_info(self):
|
| 261 |
+
"""Return status information about the speaker change detector"""
|
| 262 |
+
speaker_counts = [len(self.speaker_embeddings[i]) for i in range(self.max_speakers)]
|
| 263 |
+
|
| 264 |
+
return {
|
| 265 |
+
"current_speaker": self.current_speaker,
|
| 266 |
+
"speaker_counts": speaker_counts,
|
| 267 |
+
"active_speakers": len(self.active_speakers),
|
| 268 |
+
"max_speakers": self.max_speakers,
|
| 269 |
+
"last_similarity": self.last_similarity,
|
| 270 |
+
"threshold": self.change_threshold
|
| 271 |
+
}
|
| 272 |
|
| 273 |
|
| 274 |
+
class RealtimeSpeakerDiarization:
|
|
|
|
| 275 |
def __init__(self):
|
| 276 |
+
self.encoder = None
|
| 277 |
+
self.audio_processor = None
|
| 278 |
+
self.speaker_detector = None
|
| 279 |
+
self.recorder = None
|
| 280 |
+
self.recording_thread = None
|
| 281 |
+
self.sentence_queue = queue.Queue()
|
| 282 |
+
self.full_sentences = []
|
| 283 |
+
self.sentence_speakers = []
|
| 284 |
+
self.pending_sentences = []
|
| 285 |
+
self.displayed_text = ""
|
| 286 |
+
self.last_realtime_text = ""
|
| 287 |
+
self.is_running = False
|
| 288 |
+
self.change_threshold = DEFAULT_CHANGE_THRESHOLD
|
| 289 |
+
self.max_speakers = DEFAULT_MAX_SPEAKERS
|
| 290 |
+
|
| 291 |
+
def initialize_models(self):
|
| 292 |
+
"""Initialize the speaker encoder model"""
|
| 293 |
try:
|
| 294 |
+
device_str = "cuda" if torch.cuda.is_available() else "cpu"
|
| 295 |
+
print(f"Using device: {device_str}")
|
| 296 |
+
|
| 297 |
+
self.encoder = SpeechBrainEncoder(device=device_str)
|
| 298 |
+
success = self.encoder.load_model()
|
| 299 |
+
|
| 300 |
+
if success:
|
| 301 |
+
self.audio_processor = AudioProcessor(self.encoder)
|
| 302 |
+
self.speaker_detector = SpeakerChangeDetector(
|
| 303 |
+
embedding_dim=self.encoder.embedding_dim,
|
| 304 |
+
change_threshold=self.change_threshold,
|
| 305 |
+
max_speakers=self.max_speakers
|
| 306 |
+
)
|
| 307 |
+
print("ECAPA-TDNN model loaded successfully!")
|
| 308 |
+
return True
|
| 309 |
+
else:
|
| 310 |
+
print("Failed to load ECAPA-TDNN model")
|
| 311 |
+
return False
|
| 312 |
+
except Exception as e:
|
| 313 |
+
print(f"Model initialization error: {e}")
|
| 314 |
+
return False
|
| 315 |
+
|
| 316 |
+
def live_text_detected(self, text):
|
| 317 |
+
"""Callback for real-time transcription updates"""
|
| 318 |
+
text = text.strip()
|
| 319 |
+
if text:
|
| 320 |
+
sentence_delimiters = '.?!。'
|
| 321 |
+
prob_sentence_end = (
|
| 322 |
+
len(self.last_realtime_text) > 0
|
| 323 |
+
and text[-1] in sentence_delimiters
|
| 324 |
+
and self.last_realtime_text[-1] in sentence_delimiters
|
| 325 |
+
)
|
| 326 |
+
|
| 327 |
+
self.last_realtime_text = text
|
| 328 |
+
|
| 329 |
+
if prob_sentence_end and FAST_SENTENCE_END:
|
| 330 |
+
self.recorder.stop()
|
| 331 |
+
elif prob_sentence_end:
|
| 332 |
+
self.recorder.post_speech_silence_duration = SILENCE_THRESHS[0]
|
| 333 |
+
else:
|
| 334 |
+
self.recorder.post_speech_silence_duration = SILENCE_THRESHS[1]
|
| 335 |
+
|
| 336 |
+
def process_final_text(self, text):
|
| 337 |
+
"""Process final transcribed text with speaker embedding"""
|
| 338 |
+
text = text.strip()
|
| 339 |
+
if text:
|
| 340 |
+
try:
|
| 341 |
+
bytes_data = self.recorder.last_transcription_bytes
|
| 342 |
+
self.sentence_queue.put((text, bytes_data))
|
| 343 |
+
self.pending_sentences.append(text)
|
| 344 |
+
except Exception as e:
|
| 345 |
+
print(f"Error processing final text: {e}")
|
| 346 |
|
| 347 |
+
def process_sentence_queue(self):
|
| 348 |
+
"""Process sentences in the queue for speaker detection"""
|
| 349 |
+
while self.is_running:
|
| 350 |
+
try:
|
| 351 |
+
text, bytes_data = self.sentence_queue.get(timeout=1)
|
| 352 |
+
|
| 353 |
+
# Convert audio data to int16
|
| 354 |
+
audio_int16 = np.int16(bytes_data * 32767)
|
| 355 |
+
|
| 356 |
+
# Extract speaker embedding
|
| 357 |
+
speaker_embedding = self.audio_processor.extract_embedding(audio_int16)
|
| 358 |
+
|
| 359 |
+
# Store sentence and embedding
|
| 360 |
+
self.full_sentences.append((text, speaker_embedding))
|
| 361 |
+
|
| 362 |
+
# Fill in missing speaker assignments
|
| 363 |
+
while len(self.sentence_speakers) < len(self.full_sentences) - 1:
|
| 364 |
+
self.sentence_speakers.append(0)
|
| 365 |
+
|
| 366 |
+
# Detect speaker changes
|
| 367 |
+
speaker_id, similarity = self.speaker_detector.add_embedding(speaker_embedding)
|
| 368 |
+
self.sentence_speakers.append(speaker_id)
|
| 369 |
+
|
| 370 |
+
# Remove from pending
|
| 371 |
+
if text in self.pending_sentences:
|
| 372 |
+
self.pending_sentences.remove(text)
|
| 373 |
+
|
| 374 |
+
except queue.Empty:
|
| 375 |
+
continue
|
| 376 |
+
except Exception as e:
|
| 377 |
+
print(f"Error processing sentence: {e}")
|
| 378 |
+
|
| 379 |
+
def start_recording(self):
|
| 380 |
+
"""Start the recording and transcription process"""
|
| 381 |
+
if self.encoder is None:
|
| 382 |
+
return "Please initialize models first!"
|
| 383 |
+
|
| 384 |
try:
|
| 385 |
+
# Setup recorder configuration
|
| 386 |
+
recorder_config = {
|
| 387 |
+
'spinner': False,
|
| 388 |
+
'use_microphone': USE_MICROPHONE,
|
| 389 |
+
'model': FINAL_TRANSCRIPTION_MODEL,
|
| 390 |
+
'language': TRANSCRIPTION_LANGUAGE,
|
| 391 |
+
'silero_sensitivity': SILERO_SENSITIVITY,
|
| 392 |
+
'webrtc_sensitivity': WEBRTC_SENSITIVITY,
|
| 393 |
+
'post_speech_silence_duration': SILENCE_THRESHS[1],
|
| 394 |
+
'min_length_of_recording': MIN_LENGTH_OF_RECORDING,
|
| 395 |
+
'pre_recording_buffer_duration': PRE_RECORDING_BUFFER_DURATION,
|
| 396 |
+
'min_gap_between_recordings': 0,
|
| 397 |
+
'enable_realtime_transcription': True,
|
| 398 |
+
'realtime_processing_pause': 0,
|
| 399 |
+
'realtime_model_type': REALTIME_TRANSCRIPTION_MODEL,
|
| 400 |
+
'on_realtime_transcription_update': self.live_text_detected,
|
| 401 |
+
'beam_size': FINAL_BEAM_SIZE,
|
| 402 |
+
'beam_size_realtime': REALTIME_BEAM_SIZE,
|
| 403 |
+
'buffer_size': BUFFER_SIZE,
|
| 404 |
+
'sample_rate': SAMPLE_RATE,
|
| 405 |
+
}
|
| 406 |
+
|
| 407 |
+
self.recorder = AudioToTextRecorder(**recorder_config)
|
| 408 |
|
| 409 |
+
# Start sentence processing thread
|
| 410 |
+
self.is_running = True
|
| 411 |
+
self.sentence_thread = threading.Thread(target=self.process_sentence_queue, daemon=True)
|
| 412 |
+
self.sentence_thread.start()
|
| 413 |
|
| 414 |
+
# Start audio capture thread
|
| 415 |
+
self.audio_thread = threading.Thread(target=self.capture_audio, daemon=True)
|
| 416 |
+
self.audio_thread.start()
|
| 417 |
|
| 418 |
+
# Start transcription thread
|
| 419 |
+
self.transcription_thread = threading.Thread(target=self.run_transcription, daemon=True)
|
| 420 |
+
self.transcription_thread.start()
|
|
|
|
| 421 |
|
| 422 |
+
return "Recording started successfully!"
|
|
|
|
|
|
|
| 423 |
|
| 424 |
+
except Exception as e:
|
| 425 |
+
return f"Error starting recording: {e}"
|
| 426 |
+
|
| 427 |
+
def capture_audio(self):
|
| 428 |
+
"""Capture audio from default speaker/microphone"""
|
| 429 |
+
try:
|
| 430 |
+
device_id = str(sc.default_speaker().name if not USE_MICROPHONE else sc.default_microphone().name)
|
| 431 |
+
include_loopback = not USE_MICROPHONE
|
| 432 |
|
| 433 |
+
with sc.get_microphone(id=device_id, include_loopback=include_loopback).recorder(
|
| 434 |
+
samplerate=SAMPLE_RATE, blocksize=BUFFER_SIZE
|
| 435 |
+
) as mic:
|
| 436 |
+
while self.is_running:
|
| 437 |
+
audio_data = mic.record(numframes=BUFFER_SIZE)
|
| 438 |
+
|
| 439 |
+
if audio_data.shape[1] > 1 and CHANNELS == 1:
|
| 440 |
+
audio_data = audio_data[:, 0]
|
| 441 |
+
|
| 442 |
+
audio_int16 = (audio_data.flatten() * 32767).astype(np.int16)
|
| 443 |
+
audio_bytes = audio_int16.tobytes()
|
| 444 |
+
self.recorder.feed_audio(audio_bytes)
|
| 445 |
+
|
| 446 |
+
except Exception as e:
|
| 447 |
+
print(f"Audio capture error: {e}")
|
| 448 |
+
|
| 449 |
+
def run_transcription(self):
|
| 450 |
+
"""Run the transcription loop"""
|
| 451 |
+
try:
|
| 452 |
+
while self.is_running:
|
| 453 |
+
self.recorder.text(self.process_final_text)
|
| 454 |
except Exception as e:
|
| 455 |
print(f"Transcription error: {e}")
|
|
|
|
| 456 |
|
| 457 |
+
def stop_recording(self):
|
| 458 |
+
"""Stop the recording process"""
|
| 459 |
+
self.is_running = False
|
| 460 |
+
if self.recorder:
|
| 461 |
+
self.recorder.stop()
|
| 462 |
+
return "Recording stopped!"
|
| 463 |
|
| 464 |
+
def clear_conversation(self):
|
| 465 |
+
"""Clear all conversation data"""
|
| 466 |
+
self.full_sentences = []
|
| 467 |
+
self.sentence_speakers = []
|
| 468 |
+
self.pending_sentences = []
|
| 469 |
+
self.displayed_text = ""
|
| 470 |
+
self.last_realtime_text = ""
|
| 471 |
+
|
| 472 |
+
if self.speaker_detector:
|
| 473 |
+
self.speaker_detector = SpeakerChangeDetector(
|
| 474 |
+
embedding_dim=self.encoder.embedding_dim,
|
| 475 |
+
change_threshold=self.change_threshold,
|
| 476 |
+
max_speakers=self.max_speakers
|
| 477 |
+
)
|
|
|
|
|
|
|
| 478 |
|
| 479 |
+
return "Conversation cleared!"
|
| 480 |
|
| 481 |
+
def update_settings(self, threshold, max_speakers):
|
| 482 |
+
"""Update speaker detection settings"""
|
| 483 |
+
self.change_threshold = threshold
|
| 484 |
+
self.max_speakers = max_speakers
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 485 |
|
| 486 |
+
if self.speaker_detector:
|
| 487 |
+
self.speaker_detector.set_change_threshold(threshold)
|
| 488 |
+
self.speaker_detector.set_max_speakers(max_speakers)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 489 |
|
| 490 |
+
return f"Settings updated: Threshold={threshold:.2f}, Max Speakers={max_speakers}"
|
| 491 |
+
|
| 492 |
+
def get_formatted_conversation(self):
|
| 493 |
+
"""Get the formatted conversation with speaker colors"""
|
| 494 |
+
try:
|
| 495 |
+
sentences_with_style = []
|
| 496 |
+
|
| 497 |
+
# Process completed sentences
|
| 498 |
+
for i, sentence in enumerate(self.full_sentences):
|
| 499 |
+
sentence_text, _ = sentence
|
| 500 |
+
if i >= len(self.sentence_speakers):
|
| 501 |
+
color = "#FFFFFF"
|
| 502 |
+
else:
|
| 503 |
+
speaker_id = self.sentence_speakers[i]
|
| 504 |
+
color = self.speaker_detector.get_color_for_speaker(speaker_id)
|
| 505 |
+
speaker_name = f"Speaker {speaker_id + 1}"
|
| 506 |
+
|
| 507 |
+
sentences_with_style.append(
|
| 508 |
+
f'<span style="color:{color};"><b>{speaker_name}:</b> {sentence_text}</span>')
|
| 509 |
+
|
| 510 |
+
# Add pending sentences
|
| 511 |
+
for pending_sentence in self.pending_sentences:
|
| 512 |
+
sentences_with_style.append(
|
| 513 |
+
f'<span style="color:#60FFFF;"><b>Processing:</b> {pending_sentence}</span>')
|
| 514 |
+
|
| 515 |
+
if sentences_with_style:
|
| 516 |
+
return "<br><br>".join(sentences_with_style)
|
| 517 |
+
else:
|
| 518 |
+
return "Waiting for speech input..."
|
| 519 |
+
|
| 520 |
+
except Exception as e:
|
| 521 |
+
return f"Error formatting conversation: {e}"
|
| 522 |
|
| 523 |
def get_status_info(self):
|
| 524 |
"""Get current status information"""
|
| 525 |
+
if not self.speaker_detector:
|
| 526 |
+
return "Speaker detector not initialized"
|
| 527 |
+
|
| 528 |
+
try:
|
| 529 |
+
status = self.speaker_detector.get_status_info()
|
| 530 |
+
|
| 531 |
+
status_lines = [
|
| 532 |
+
f"**Current Speaker:** {status['current_speaker'] + 1}",
|
| 533 |
+
f"**Active Speakers:** {status['active_speakers']} of {status['max_speakers']}",
|
| 534 |
+
f"**Last Similarity:** {status['last_similarity']:.3f}",
|
| 535 |
+
f"**Change Threshold:** {status['threshold']:.2f}",
|
| 536 |
+
f"**Total Sentences:** {len(self.full_sentences)}",
|
| 537 |
+
"",
|
| 538 |
+
"**Speaker Segment Counts:**"
|
| 539 |
+
]
|
| 540 |
+
|
| 541 |
+
for i in range(status['max_speakers']):
|
| 542 |
+
color_name = SPEAKER_COLOR_NAMES[i] if i < len(SPEAKER_COLOR_NAMES) else f"Speaker {i+1}"
|
| 543 |
+
status_lines.append(f"Speaker {i+1} ({color_name}): {status['speaker_counts'][i]}")
|
| 544 |
+
|
| 545 |
+
return "\n".join(status_lines)
|
| 546 |
+
|
| 547 |
+
except Exception as e:
|
| 548 |
+
return f"Error getting status: {e}"
|
| 549 |
|
| 550 |
|
| 551 |
# Global instance
|
| 552 |
+
diarization_system = RealtimeSpeakerDiarization()
|
| 553 |
|
| 554 |
|
| 555 |
+
def initialize_system():
|
| 556 |
+
"""Initialize the diarization system"""
|
| 557 |
+
success = diarization_system.initialize_models()
|
| 558 |
+
if success:
|
| 559 |
+
return "✅ System initialized successfully! Models loaded."
|
| 560 |
+
else:
|
| 561 |
+
return "❌ Failed to initialize system. Please check the logs."
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 562 |
|
| 563 |
|
| 564 |
+
def start_recording():
|
| 565 |
+
"""Start recording and transcription"""
|
| 566 |
+
return diarization_system.start_recording()
|
| 567 |
+
|
| 568 |
+
|
| 569 |
+
def stop_recording():
|
| 570 |
+
"""Stop recording and transcription"""
|
| 571 |
+
return diarization_system.stop_recording()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 572 |
|
| 573 |
|
| 574 |
def clear_conversation():
|
| 575 |
"""Clear the conversation"""
|
| 576 |
+
return diarization_system.clear_conversation()
|
| 577 |
+
|
| 578 |
+
|
| 579 |
+
def update_settings(threshold, max_speakers):
|
| 580 |
+
"""Update system settings"""
|
| 581 |
+
return diarization_system.update_settings(threshold, max_speakers)
|
| 582 |
+
|
| 583 |
|
| 584 |
+
def get_conversation():
|
| 585 |
+
"""Get the current conversation"""
|
| 586 |
+
return diarization_system.get_formatted_conversation()
|
| 587 |
|
| 588 |
+
|
| 589 |
+
def get_status():
|
| 590 |
+
"""Get system status"""
|
| 591 |
+
return diarization_system.get_status_info()
|
| 592 |
+
|
| 593 |
+
|
| 594 |
+
# Create Gradio interface
|
| 595 |
def create_interface():
|
| 596 |
+
with gr.Blocks(title="Real-time Speaker Diarization", theme=gr.themes.Dark()) as app:
|
| 597 |
+
gr.Markdown("# 🎤 Real-time Speech Recognition with Speaker Diarization")
|
| 598 |
+
gr.Markdown("This app performs real-time speech recognition with automatic speaker identification and color-coding.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 599 |
|
| 600 |
with gr.Row():
|
| 601 |
+
with gr.Column(scale=2):
|
| 602 |
# Main conversation display
|
| 603 |
+
conversation_output = gr.HTML(
|
| 604 |
+
value="<i>Click 'Initialize System' to start...</i>",
|
| 605 |
+
label="Live Conversation"
|
| 606 |
)
|
| 607 |
|
| 608 |
+
# Control buttons
|
| 609 |
+
with gr.Row():
|
| 610 |
+
init_btn = gr.Button("🔧 Initialize System", variant="secondary")
|
| 611 |
+
start_btn = gr.Button("🎙️ Start Recording", variant="primary", interactive=False)
|
| 612 |
+
stop_btn = gr.Button("⏹️ Stop Recording", variant="stop", interactive=False)
|
| 613 |
+
clear_btn = gr.Button("🗑️ Clear Conversation", interactive=False)
|
|
|
|
| 614 |
|
| 615 |
+
# Status display
|
| 616 |
+
status_output = gr.Textbox(
|
| 617 |
+
label="System Status",
|
| 618 |
+
value="System not initialized",
|
| 619 |
+
lines=8,
|
| 620 |
+
interactive=False
|
| 621 |
+
)
|
| 622 |
+
|
| 623 |
with gr.Column(scale=1):
|
| 624 |
+
# Settings panel
|
| 625 |
+
gr.Markdown("## ⚙️ Settings")
|
| 626 |
|
| 627 |
threshold_slider = gr.Slider(
|
| 628 |
minimum=0.1,
|
| 629 |
+
maximum=0.95,
|
|
|
|
| 630 |
step=0.05,
|
| 631 |
+
value=DEFAULT_CHANGE_THRESHOLD,
|
| 632 |
+
label="Speaker Change Sensitivity",
|
| 633 |
+
info="Lower values = more sensitive to speaker changes"
|
| 634 |
)
|
| 635 |
|
| 636 |
max_speakers_slider = gr.Slider(
|
| 637 |
minimum=2,
|
| 638 |
maximum=ABSOLUTE_MAX_SPEAKERS,
|
|
|
|
| 639 |
step=1,
|
| 640 |
+
value=DEFAULT_MAX_SPEAKERS,
|
| 641 |
+
label="Maximum Number of Speakers"
|
| 642 |
)
|
| 643 |
|
| 644 |
+
update_settings_btn = gr.Button("Update Settings")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 645 |
|
| 646 |
# Speaker color legend
|
| 647 |
+
gr.Markdown("## 🎨 Speaker Colors")
|
| 648 |
+
color_info = []
|
| 649 |
+
for i, (color, name) in enumerate(zip(SPEAKER_COLORS, SPEAKER_COLOR_NAMES)):
|
| 650 |
+
color_info.append(f'<span style="color:{color};">■</span> Speaker {i+1} ({name})')
|
|
|
|
|
|
|
| 651 |
|
| 652 |
+
gr.HTML("<br>".join(color_info[:DEFAULT_MAX_SPEAKERS]))
|
| 653 |
+
|
| 654 |
+
# Auto-refresh conversation and status
|
| 655 |
+
def refresh_display():
|
| 656 |
+
return get_conversation(), get_status()
|
| 657 |
|
| 658 |
# Event handlers
|
| 659 |
+
def on_initialize():
|
| 660 |
+
result = initialize_system()
|
| 661 |
+
if "successfully" in result:
|
| 662 |
+
return (
|
| 663 |
+
result,
|
| 664 |
+
gr.update(interactive=True), # start_btn
|
| 665 |
+
gr.update(interactive=True), # clear_btn
|
| 666 |
+
get_conversation(),
|
| 667 |
+
get_status()
|
| 668 |
+
)
|
| 669 |
+
else:
|
| 670 |
+
return (
|
| 671 |
+
result,
|
| 672 |
+
gr.update(interactive=False), # start_btn
|
| 673 |
+
gr.update(interactive=False), # clear_btn
|
| 674 |
+
get_conversation(),
|
| 675 |
+
get_status()
|
| 676 |
+
)
|
| 677 |
+
|
| 678 |
+
def on_start():
|
| 679 |
+
result = start_recording()
|
| 680 |
+
return (
|
| 681 |
+
result,
|
| 682 |
+
gr.update(interactive=False), # start_btn
|
| 683 |
+
gr.update(interactive=True), # stop_btn
|
| 684 |
+
)
|
| 685 |
+
|
| 686 |
+
def on_stop():
|
| 687 |
+
result = stop_recording()
|
| 688 |
+
return (
|
| 689 |
+
result,
|
| 690 |
+
gr.update(interactive=True), # start_btn
|
| 691 |
+
gr.update(interactive=False), # stop_btn
|
| 692 |
+
)
|
| 693 |
+
|
| 694 |
+
# Connect event handlers
|
| 695 |
+
init_btn.click(
|
| 696 |
+
on_initialize,
|
| 697 |
+
outputs=[status_output, start_btn, clear_btn, conversation_output, status_output]
|
| 698 |
+
)
|
| 699 |
+
|
| 700 |
+
start_btn.click(
|
| 701 |
+
on_start,
|
| 702 |
+
outputs=[status_output, start_btn, stop_btn]
|
| 703 |
+
)
|
| 704 |
+
|
| 705 |
+
stop_btn.click(
|
| 706 |
+
on_stop,
|
| 707 |
+
outputs=[status_output, start_btn, stop_btn]
|
| 708 |
)
|
| 709 |
|
| 710 |
clear_btn.click(
|
| 711 |
+
clear_conversation,
|
| 712 |
+
outputs=[status_output]
|
| 713 |
+
)
|
| 714 |
+
|
| 715 |
+
update_settings_btn.click(
|
| 716 |
+
update_settings,
|
| 717 |
+
inputs=[threshold_slider, max_speakers_slider],
|
| 718 |
+
outputs=[status_output]
|
| 719 |
)
|
| 720 |
|
| 721 |
+
# Auto-refresh every 2 seconds when recording
|
| 722 |
+
refresh_timer = gr.Timer(2.0)
|
| 723 |
+
refresh_timer.tick(
|
| 724 |
+
refresh_display,
|
| 725 |
+
outputs=[conversation_output, status_output]
|
| 726 |
)
|
| 727 |
|
| 728 |
+
return app
|
| 729 |
|
| 730 |
|
| 731 |
if __name__ == "__main__":
|
| 732 |
+
app = create_interface()
|
| 733 |
+
app.launch(
|
|
|
|
| 734 |
server_name="0.0.0.0",
|
| 735 |
server_port=7860,
|
| 736 |
share=True
|