Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -12,449 +12,301 @@ import time
|
|
| 12 |
|
| 13 |
# تنظیمات اولیه
|
| 14 |
recognizer = sr.Recognizer()
|
| 15 |
-
recognizer.energy_threshold =
|
| 16 |
recognizer.dynamic_energy_threshold = True
|
| 17 |
-
recognizer.pause_threshold = 0.
|
| 18 |
|
| 19 |
-
# صف برای accumulate
|
| 20 |
-
audio_queue = queue.Queue(maxsize=
|
| 21 |
-
|
| 22 |
-
|
| 23 |
|
| 24 |
-
#
|
| 25 |
current_transcript = ""
|
| 26 |
transcript_lock = threading.Lock()
|
| 27 |
|
| 28 |
-
#
|
| 29 |
def background_processor():
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
last_process_time = time.time()
|
| 33 |
|
| 34 |
while True:
|
| 35 |
try:
|
| 36 |
-
#
|
| 37 |
if not audio_queue.empty():
|
| 38 |
-
audio_tuple = audio_queue.get(timeout=0.
|
| 39 |
-
if audio_tuple:
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
audio_data = np.clip(audio_data, -1.0, 1.0)
|
| 45 |
-
accumulated_audio.append((sample_rate, audio_data))
|
| 46 |
-
print(f"Chunk accumulated: {len(audio_data)/sample_rate:.2f}s") # log
|
| 47 |
|
| 48 |
-
#
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
if (
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
merged_data = np.concatenate([data for _, data in accumulated_audio])
|
| 56 |
|
| 57 |
text = process_audio_chunk((merged_rate, merged_data))
|
| 58 |
-
if text and text
|
| 59 |
with transcript_lock:
|
| 60 |
if current_transcript:
|
| 61 |
current_transcript += " " + text
|
| 62 |
else:
|
| 63 |
current_transcript = text
|
| 64 |
-
print(f"Processed: {text}") #
|
| 65 |
|
| 66 |
-
# Reset
|
| 67 |
-
|
| 68 |
-
|
| 69 |
|
| 70 |
-
time.sleep(0.
|
| 71 |
|
| 72 |
-
except queue.Empty:
|
| 73 |
-
continue
|
| 74 |
except Exception as e:
|
| 75 |
-
print(f"
|
| 76 |
-
time.sleep(
|
| 77 |
|
| 78 |
-
#
|
| 79 |
processor_thread = threading.Thread(target=background_processor, daemon=True)
|
| 80 |
processor_thread.start()
|
| 81 |
|
| 82 |
def numpy_to_audio_segment(audio_data, sample_rate):
|
| 83 |
-
"""تبدیل numpy array به AudioSegment (بهبود: clip و handling بهتر)"""
|
| 84 |
if audio_data is None or len(audio_data) == 0:
|
| 85 |
return None
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
if audio_data.dtype == np.float32 or audio_data.dtype == np.float64:
|
| 89 |
-
audio_data = np.clip(audio_data, -1.0, 1.0) # clip برای جلوگیری از distortion
|
| 90 |
audio_data = (audio_data * 32767).astype(np.int16)
|
| 91 |
|
| 92 |
-
# ذخیره در buffer موقت
|
| 93 |
buffer = io.BytesIO()
|
| 94 |
with wave.open(buffer, 'wb') as wav_file:
|
| 95 |
wav_file.setnchannels(1)
|
| 96 |
wav_file.setsampwidth(2)
|
| 97 |
wav_file.setframerate(sample_rate)
|
| 98 |
wav_file.writeframes(audio_data.tobytes())
|
| 99 |
-
|
| 100 |
buffer.seek(0)
|
| 101 |
return AudioSegment.from_wav(buffer)
|
| 102 |
|
| 103 |
def process_audio_chunk(audio_tuple):
|
| 104 |
-
"""پردازش یک قطعه صوتی با Google Speech Recognition (بهبود: handling chunk کوتاه)"""
|
| 105 |
try:
|
| 106 |
if audio_tuple is None:
|
| 107 |
return ""
|
| 108 |
-
|
| 109 |
sample_rate, audio_data = audio_tuple
|
| 110 |
-
duration = len(audio_data) / sample_rate if sample_rate
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
return "" # skip chunkهای خیلی کوتاه
|
| 114 |
|
| 115 |
-
# تبدیل به AudioSegment
|
| 116 |
audio_segment = numpy_to_audio_segment(audio_data, sample_rate)
|
| 117 |
-
if audio_segment is None or len(audio_segment) <
|
| 118 |
return ""
|
| 119 |
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
tmp_path = tmp_file.name
|
| 124 |
|
| 125 |
-
# استفاده از speech recognition با adjust برای chunk
|
| 126 |
with sr.AudioFile(tmp_path) as source:
|
| 127 |
-
recognizer.adjust_for_ambient_noise(source, duration=0.
|
| 128 |
audio = recognizer.record(source)
|
| 129 |
-
|
| 130 |
-
#
|
| 131 |
text = ""
|
| 132 |
try:
|
| 133 |
-
text = recognizer.recognize_google(audio, language='fa-IR')
|
|
|
|
| 134 |
except sr.UnknownValueError:
|
| 135 |
-
# تلاش با انگلیسی
|
| 136 |
try:
|
| 137 |
text = recognizer.recognize_google(audio, language='en-US')
|
| 138 |
except sr.UnknownValueError:
|
| 139 |
-
|
| 140 |
-
|
| 141 |
except sr.RequestError as e:
|
| 142 |
-
text = "[خطا در اتصال]"
|
| 143 |
print(f"Google API error: {e}")
|
|
|
|
| 144 |
|
| 145 |
-
# پاک کردن فایل موقت
|
| 146 |
if os.path.exists(tmp_path):
|
| 147 |
os.unlink(tmp_path)
|
| 148 |
-
|
| 149 |
-
return text.strip()
|
| 150 |
|
|
|
|
| 151 |
except Exception as e:
|
| 152 |
-
print(f"
|
| 153 |
return ""
|
| 154 |
|
| 155 |
-
def
|
| 156 |
-
"""
|
| 157 |
-
global current_transcript
|
| 158 |
-
|
| 159 |
if audio is None:
|
| 160 |
-
return
|
| 161 |
-
|
| 162 |
-
# Add to queue (non-blocking)
|
| 163 |
try:
|
| 164 |
audio_queue.put(audio, block=False)
|
| 165 |
except queue.Full:
|
| 166 |
-
print("Queue full,
|
| 167 |
-
|
| 168 |
-
|
|
|
|
|
|
|
| 169 |
with transcript_lock:
|
| 170 |
return current_transcript
|
| 171 |
|
| 172 |
-
def
|
| 173 |
-
"""پاک کردن متن real-time"""
|
| 174 |
global current_transcript
|
| 175 |
with transcript_lock:
|
| 176 |
current_transcript = ""
|
| 177 |
-
# Clear queue
|
| 178 |
while not audio_queue.empty():
|
| 179 |
-
|
| 180 |
-
audio_queue.get_nowait()
|
| 181 |
-
except queue.Empty:
|
| 182 |
-
break
|
| 183 |
return ""
|
| 184 |
|
| 185 |
-
#
|
| 186 |
def transcribe_file(audio_file, chunk_duration=30):
|
| 187 |
-
"""تبدیل فایل صوتی به متن با تقسیم به بخشها"""
|
| 188 |
if audio_file is None:
|
| 189 |
-
yield "لطفاً
|
| 190 |
return
|
| 191 |
-
|
| 192 |
try:
|
| 193 |
-
# خواندن فایل صوتی
|
| 194 |
audio = AudioSegment.from_file(audio_file)
|
| 195 |
duration_ms = len(audio)
|
| 196 |
-
|
| 197 |
-
|
| 198 |
all_text = []
|
| 199 |
-
progress_text = ""
|
| 200 |
|
| 201 |
-
|
| 202 |
-
|
| 203 |
-
|
| 204 |
-
|
| 205 |
-
|
| 206 |
-
|
| 207 |
-
# ذخیره chunk موقت
|
| 208 |
-
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp_file:
|
| 209 |
-
chunk.export(tmp_file.name, format="wav")
|
| 210 |
-
tmp_path = tmp_file.name
|
| 211 |
|
| 212 |
-
# تشخیص گفتار
|
| 213 |
try:
|
| 214 |
with sr.AudioFile(tmp_path) as source:
|
| 215 |
audio_data = recognizer.record(source)
|
| 216 |
-
|
| 217 |
-
# تلاش با فارسی
|
| 218 |
try:
|
| 219 |
text = recognizer.recognize_google(audio_data, language='fa-IR')
|
| 220 |
except sr.UnknownValueError:
|
| 221 |
-
# تلاش با انگلیسی
|
| 222 |
try:
|
| 223 |
text = recognizer.recognize_google(audio_data, language='en-US')
|
| 224 |
except sr.UnknownValueError:
|
| 225 |
text = ""
|
| 226 |
except sr.RequestError:
|
| 227 |
-
text = "[خطا
|
| 228 |
-
|
| 229 |
-
if text and text != "[خطا در اتصال به سرویس]":
|
| 230 |
all_text.append(text)
|
| 231 |
-
|
| 232 |
except Exception as e:
|
| 233 |
-
print(f"
|
| 234 |
|
| 235 |
-
# پاک کردن فایل موقت
|
| 236 |
if os.path.exists(tmp_path):
|
| 237 |
os.unlink(tmp_path)
|
| 238 |
|
| 239 |
-
|
| 240 |
-
|
| 241 |
-
progress = min((i + chunk_duration_ms) / duration_ms * 100, 100)
|
| 242 |
-
progress_text = f"پیشرفت: {progress:.1f}% - بخش {current_chunk} از {num_chunks}"
|
| 243 |
-
|
| 244 |
-
yield " ".join(all_text), progress_text
|
| 245 |
-
|
| 246 |
-
# تاخیر کوتاه برای جلوگیری از rate limiting
|
| 247 |
time.sleep(0.5)
|
| 248 |
|
| 249 |
-
|
| 250 |
-
yield final_text, "تبدیل کامل شد! ✅"
|
| 251 |
-
|
| 252 |
except Exception as e:
|
| 253 |
-
yield f"خطا: {
|
| 254 |
|
| 255 |
def save_text(text):
|
| 256 |
-
"""ذخیره متن در فایل موقت و برگرداندن آن"""
|
| 257 |
if not text.strip():
|
| 258 |
return None
|
| 259 |
-
|
| 260 |
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt', encoding='utf-8')
|
| 261 |
temp_file.write(text)
|
| 262 |
temp_file.close()
|
| 263 |
-
|
| 264 |
return temp_file.name
|
| 265 |
|
| 266 |
-
#
|
| 267 |
with gr.Blocks(
|
| 268 |
-
title="تبدیل گفتار به متن",
|
| 269 |
theme=gr.themes.Soft(),
|
| 270 |
css="""
|
| 271 |
-
.gradio-container {
|
| 272 |
-
|
| 273 |
-
}
|
| 274 |
-
.rtl {
|
| 275 |
-
direction: rtl;
|
| 276 |
-
text-align: right;
|
| 277 |
-
}
|
| 278 |
"""
|
| 279 |
) as demo:
|
| 280 |
gr.HTML("""
|
| 281 |
<div style="text-align: center; max-width: 800px; margin: 0 auto;">
|
| 282 |
-
<h1 style="font-size: 2.5em; margin-bottom: 0.5em;">
|
| 283 |
-
🎤 تبدیل گفتار به متن
|
| 284 |
-
</h1>
|
| 285 |
<p style="font-size: 1.1em; color: #666; margin-bottom: 2em;">
|
| 286 |
-
|
| 287 |
</p>
|
| 288 |
</div>
|
| 289 |
""")
|
| 290 |
|
| 291 |
with gr.Tabs():
|
| 292 |
-
# تب Real-time
|
| 293 |
with gr.TabItem("🎙️ ضبط مستقیم"):
|
| 294 |
-
gr.Markdown("###
|
| 295 |
|
| 296 |
with gr.Row():
|
| 297 |
-
|
| 298 |
-
|
| 299 |
-
|
| 300 |
-
|
| 301 |
-
|
| 302 |
-
|
| 303 |
-
|
| 304 |
-
|
| 305 |
-
|
| 306 |
-
|
| 307 |
-
with gr.Row():
|
| 308 |
-
clear_btn = gr.Button(
|
| 309 |
-
"🗑️ پاک کردن متن",
|
| 310 |
-
variant="secondary",
|
| 311 |
-
size="sm"
|
| 312 |
-
)
|
| 313 |
|
| 314 |
realtime_output = gr.Textbox(
|
| 315 |
-
label="متن
|
| 316 |
-
placeholder="
|
| 317 |
-
lines=
|
| 318 |
elem_classes="rtl",
|
| 319 |
rtl=True,
|
| 320 |
-
show_copy_button=True
|
| 321 |
-
|
| 322 |
-
|
| 323 |
-
# اتصال برای real-time transcription (بهبود: stream هر chunk به queue)
|
| 324 |
-
audio_input.stream(
|
| 325 |
-
transcribe_realtime,
|
| 326 |
-
inputs=[audio_input],
|
| 327 |
-
outputs=[realtime_output],
|
| 328 |
-
show_progress=False
|
| 329 |
)
|
| 330 |
|
| 331 |
-
clear_btn.
|
| 332 |
-
clear_realtime,
|
| 333 |
-
outputs=[realtime_output]
|
| 334 |
-
)
|
| 335 |
|
| 336 |
-
|
| 337 |
-
|
| 338 |
-
gr.Markdown("### فایل صوتی خود را انتخاب کنید", elem_classes="rtl")
|
| 339 |
|
| 340 |
-
|
| 341 |
-
|
| 342 |
-
|
| 343 |
-
sources=["upload"],
|
| 344 |
-
type="filepath",
|
| 345 |
-
label="انتخاب فایل صوتی",
|
| 346 |
-
elem_classes="rtl"
|
| 347 |
-
)
|
| 348 |
-
|
| 349 |
-
with gr.Column(scale=1):
|
| 350 |
-
chunk_duration = gr.Slider(
|
| 351 |
-
minimum=10,
|
| 352 |
-
maximum=60,
|
| 353 |
-
value=30,
|
| 354 |
-
step=5,
|
| 355 |
-
label="مدت هر بخش (ثانیه)",
|
| 356 |
-
elem_classes="rtl"
|
| 357 |
-
)
|
| 358 |
|
| 359 |
-
|
| 360 |
-
"🚀 شروع تبدیل",
|
| 361 |
-
variant="primary",
|
| 362 |
-
size="lg"
|
| 363 |
-
)
|
| 364 |
|
| 365 |
-
|
| 366 |
-
|
| 367 |
-
|
| 368 |
-
elem_classes="rtl"
|
| 369 |
-
)
|
| 370 |
|
| 371 |
-
|
| 372 |
-
|
| 373 |
-
placeholder="متن پس از پردازش اینجا نمایش داده میشود...",
|
| 374 |
-
lines=12,
|
| 375 |
-
elem_classes="rtl",
|
| 376 |
-
rtl=True,
|
| 377 |
-
show_copy_button=True
|
| 378 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 379 |
|
| 380 |
-
# دکمههای ذخیره و پاک کردن
|
| 381 |
with gr.Row():
|
| 382 |
-
|
| 383 |
-
|
| 384 |
-
variant="secondary"
|
| 385 |
-
)
|
| 386 |
-
clear_file_btn = gr.Button(
|
| 387 |
-
"🗑️ پاک کردن",
|
| 388 |
-
variant="secondary"
|
| 389 |
-
)
|
| 390 |
-
download_file = gr.File(
|
| 391 |
-
label="دانلود فایل متن",
|
| 392 |
-
visible=False,
|
| 393 |
-
elem_classes="rtl"
|
| 394 |
-
)
|
| 395 |
|
| 396 |
-
|
| 397 |
-
|
| 398 |
-
|
| 399 |
-
inputs=[file_input, chunk_duration],
|
| 400 |
-
outputs=[file_output, progress_label]
|
| 401 |
-
)
|
| 402 |
|
| 403 |
-
|
| 404 |
-
|
| 405 |
-
|
| 406 |
-
|
| 407 |
-
).then(
|
| 408 |
-
lambda: gr.update(visible=True),
|
| 409 |
-
outputs=[download_file]
|
| 410 |
-
)
|
| 411 |
|
| 412 |
-
|
| 413 |
-
|
| 414 |
-
|
| 415 |
-
)
|
| 416 |
|
| 417 |
-
|
| 418 |
-
with gr.Accordion("📖 راهنمای استفاده", open=False, elem_classes="rtl"):
|
| 419 |
gr.Markdown("""
|
| 420 |
-
###
|
| 421 |
-
|
| 422 |
-
|
| 423 |
-
|
| 424 |
-
|
| 425 |
-
|
| 426 |
-
|
| 427 |
-
|
| 428 |
-
|
| 429 |
-
1. به تب "فایل صوتی" بروید
|
| 430 |
-
2. فایل مورد نظر را انتخاب کنید
|
| 431 |
-
3. مدت زمان تقسیمبندی را تنظیم کنید (پیشفرض: 30 ثانیه)
|
| 432 |
-
4. روی "شروع تبدیل" کلیک کنید
|
| 433 |
-
5. منتظر بمانید تا پردازش کامل شود
|
| 434 |
-
|
| 435 |
-
### فرمتهای پشتیبانی شده:
|
| 436 |
-
- MP3, WAV, M4A, FLAC, OGG, MP4, AVI, MOV
|
| 437 |
-
|
| 438 |
-
### نکات مهم:
|
| 439 |
-
- 🎯 برای دقت بیشتر، از فایلهای با کیفیت بالا استفاده کنید
|
| 440 |
-
- 🔇 نویز پسزمینه را به حداقل برسانید
|
| 441 |
-
- 🗣️ واضح و شمرده صحبت کنید (حداقل 3 ثانیه)
|
| 442 |
-
- 🌐 اتصال اینترنت پایدار داشته باشید
|
| 443 |
-
- ⚠️ Real-time ممکن است 1-2 ثانیه تاخیر داشته باشد
|
| 444 |
""", elem_classes="rtl")
|
| 445 |
|
| 446 |
-
#
|
| 447 |
-
gr.HTML("""
|
| 448 |
-
<div style="text-align: center; margin-top: 2em; padding: 1em; background-color: #f8f9fa;">
|
| 449 |
-
<p style="color: #666;">
|
| 450 |
-
ساخته شده با ❤️ | نسخه 2.1 (بهبود real-time)
|
| 451 |
-
</p>
|
| 452 |
-
</div>
|
| 453 |
-
""")
|
| 454 |
|
| 455 |
-
# اجرای برنامه
|
| 456 |
if __name__ == "__main__":
|
| 457 |
-
demo.queue().launch(
|
| 458 |
-
share=True,
|
| 459 |
-
show_error=True
|
| 460 |
-
)
|
|
|
|
| 12 |
|
| 13 |
# تنظیمات اولیه
|
| 14 |
recognizer = sr.Recognizer()
|
| 15 |
+
recognizer.energy_threshold = 400 # کمی بالاتر برای real-time
|
| 16 |
recognizer.dynamic_energy_threshold = True
|
| 17 |
+
recognizer.pause_threshold = 0.5 # حساستر به pause
|
| 18 |
|
| 19 |
+
# صف برای accumulate chunkهای real-time
|
| 20 |
+
audio_queue = queue.Queue(maxsize=10) # کوچکتر برای real-time
|
| 21 |
+
PROCESS_INTERVAL = 3.0 # هر 3 ثانیه process
|
| 22 |
+
MIN_DURATION = 1.5 # حداقل 1.5s صوت برای process
|
| 23 |
|
| 24 |
+
# Transcript global
|
| 25 |
current_transcript = ""
|
| 26 |
transcript_lock = threading.Lock()
|
| 27 |
|
| 28 |
+
# Background thread برای پردازش queue
|
| 29 |
def background_processor():
|
| 30 |
+
accumulated = []
|
| 31 |
+
last_process = time.time()
|
|
|
|
| 32 |
|
| 33 |
while True:
|
| 34 |
try:
|
| 35 |
+
# Get chunk if available
|
| 36 |
if not audio_queue.empty():
|
| 37 |
+
audio_tuple = audio_queue.get(timeout=0.2)
|
| 38 |
+
if audio_tuple and audio_tuple[1] is not None:
|
| 39 |
+
rate, data = audio_tuple
|
| 40 |
+
data = np.clip(data, -1.0, 1.0) # Normalize
|
| 41 |
+
accumulated.append((rate, data))
|
| 42 |
+
print(f"Accumulated chunk: {len(data)/rate:.2f}s")
|
|
|
|
|
|
|
|
|
|
| 43 |
|
| 44 |
+
# Process if interval passed and enough audio
|
| 45 |
+
now = time.time()
|
| 46 |
+
total_dur = sum(len(d)/r for r, d in accumulated) if accumulated else 0
|
| 47 |
+
if (now - last_process >= PROCESS_INTERVAL) and total_dur >= MIN_DURATION:
|
| 48 |
+
if accumulated:
|
| 49 |
+
merged_rate = accumulated[0][0]
|
| 50 |
+
merged_data = np.concatenate([d for _, d in accumulated])
|
|
|
|
| 51 |
|
| 52 |
text = process_audio_chunk((merged_rate, merged_data))
|
| 53 |
+
if text and text not in ["", "[خطا در اتصال]"]:
|
| 54 |
with transcript_lock:
|
| 55 |
if current_transcript:
|
| 56 |
current_transcript += " " + text
|
| 57 |
else:
|
| 58 |
current_transcript = text
|
| 59 |
+
print(f"✅ Processed: {text[:50]}...") # Log کوتاه
|
| 60 |
|
| 61 |
+
# Reset
|
| 62 |
+
accumulated = []
|
| 63 |
+
last_process = now
|
| 64 |
|
| 65 |
+
time.sleep(0.2) # Poll faster for responsiveness
|
| 66 |
|
|
|
|
|
|
|
| 67 |
except Exception as e:
|
| 68 |
+
print(f"Background error: {e}")
|
| 69 |
+
time.sleep(1)
|
| 70 |
|
| 71 |
+
# Start background thread
|
| 72 |
processor_thread = threading.Thread(target=background_processor, daemon=True)
|
| 73 |
processor_thread.start()
|
| 74 |
|
| 75 |
def numpy_to_audio_segment(audio_data, sample_rate):
|
|
|
|
| 76 |
if audio_data is None or len(audio_data) == 0:
|
| 77 |
return None
|
| 78 |
+
if audio_data.dtype in [np.float32, np.float64]:
|
| 79 |
+
audio_data = np.clip(audio_data, -1.0, 1.0)
|
|
|
|
|
|
|
| 80 |
audio_data = (audio_data * 32767).astype(np.int16)
|
| 81 |
|
|
|
|
| 82 |
buffer = io.BytesIO()
|
| 83 |
with wave.open(buffer, 'wb') as wav_file:
|
| 84 |
wav_file.setnchannels(1)
|
| 85 |
wav_file.setsampwidth(2)
|
| 86 |
wav_file.setframerate(sample_rate)
|
| 87 |
wav_file.writeframes(audio_data.tobytes())
|
|
|
|
| 88 |
buffer.seek(0)
|
| 89 |
return AudioSegment.from_wav(buffer)
|
| 90 |
|
| 91 |
def process_audio_chunk(audio_tuple):
|
|
|
|
| 92 |
try:
|
| 93 |
if audio_tuple is None:
|
| 94 |
return ""
|
|
|
|
| 95 |
sample_rate, audio_data = audio_tuple
|
| 96 |
+
duration = len(audio_data) / sample_rate if sample_rate else 0
|
| 97 |
+
if duration < MIN_DURATION:
|
| 98 |
+
return ""
|
|
|
|
| 99 |
|
|
|
|
| 100 |
audio_segment = numpy_to_audio_segment(audio_data, sample_rate)
|
| 101 |
+
if audio_segment is None or len(audio_segment) < MIN_DURATION * 1000:
|
| 102 |
return ""
|
| 103 |
|
| 104 |
+
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
|
| 105 |
+
audio_segment.export(tmp.name, format="wav")
|
| 106 |
+
tmp_path = tmp.name
|
|
|
|
| 107 |
|
|
|
|
| 108 |
with sr.AudioFile(tmp_path) as source:
|
| 109 |
+
recognizer.adjust_for_ambient_noise(source, duration=0.3) # Quick adjust
|
| 110 |
audio = recognizer.record(source)
|
| 111 |
+
|
| 112 |
+
# Google recognition
|
| 113 |
text = ""
|
| 114 |
try:
|
| 115 |
+
text = recognizer.recognize_google(audio, language='fa-IR') # Persian first
|
| 116 |
+
# اگر key داری: recognize_google(audio, language='fa-IR', key="YOUR_GOOGLE_API_KEY")
|
| 117 |
except sr.UnknownValueError:
|
|
|
|
| 118 |
try:
|
| 119 |
text = recognizer.recognize_google(audio, language='en-US')
|
| 120 |
except sr.UnknownValueError:
|
| 121 |
+
print("No speech in chunk")
|
| 122 |
+
text = ""
|
| 123 |
except sr.RequestError as e:
|
|
|
|
| 124 |
print(f"Google API error: {e}")
|
| 125 |
+
text = "[خطا اتصال]"
|
| 126 |
|
|
|
|
| 127 |
if os.path.exists(tmp_path):
|
| 128 |
os.unlink(tmp_path)
|
|
|
|
|
|
|
| 129 |
|
| 130 |
+
return text.strip()
|
| 131 |
except Exception as e:
|
| 132 |
+
print(f"Process error: {e}")
|
| 133 |
return ""
|
| 134 |
|
| 135 |
+
def handle_realtime_audio(audio):
|
| 136 |
+
"""Handle incoming audio chunks from microphone"""
|
|
|
|
|
|
|
| 137 |
if audio is None:
|
| 138 |
+
return gr.update()
|
|
|
|
|
|
|
| 139 |
try:
|
| 140 |
audio_queue.put(audio, block=False)
|
| 141 |
except queue.Full:
|
| 142 |
+
print("Queue full, skip")
|
| 143 |
+
return gr.update()
|
| 144 |
+
|
| 145 |
+
def get_current_transcript():
|
| 146 |
+
"""Poll transcript for UI update"""
|
| 147 |
with transcript_lock:
|
| 148 |
return current_transcript
|
| 149 |
|
| 150 |
+
def clear_transcript():
|
|
|
|
| 151 |
global current_transcript
|
| 152 |
with transcript_lock:
|
| 153 |
current_transcript = ""
|
| 154 |
+
# Clear queue
|
| 155 |
while not audio_queue.empty():
|
| 156 |
+
audio_queue.get_nowait()
|
|
|
|
|
|
|
|
|
|
| 157 |
return ""
|
| 158 |
|
| 159 |
+
# File transcription (unchanged, works fine)
|
| 160 |
def transcribe_file(audio_file, chunk_duration=30):
|
|
|
|
| 161 |
if audio_file is None:
|
| 162 |
+
yield "لطفاً فایل آپلود کنید", ""
|
| 163 |
return
|
|
|
|
| 164 |
try:
|
|
|
|
| 165 |
audio = AudioSegment.from_file(audio_file)
|
| 166 |
duration_ms = len(audio)
|
| 167 |
+
chunk_ms = chunk_duration * 1000
|
|
|
|
| 168 |
all_text = []
|
|
|
|
| 169 |
|
| 170 |
+
num_chunks = (duration_ms + chunk_ms - 1) // chunk_ms
|
| 171 |
+
for i in range(0, duration_ms, chunk_ms):
|
| 172 |
+
chunk = audio[i:i + chunk_ms]
|
| 173 |
+
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
|
| 174 |
+
chunk.export(tmp.name, format="wav")
|
| 175 |
+
tmp_path = tmp.name
|
|
|
|
|
|
|
|
|
|
|
|
|
| 176 |
|
|
|
|
| 177 |
try:
|
| 178 |
with sr.AudioFile(tmp_path) as source:
|
| 179 |
audio_data = recognizer.record(source)
|
|
|
|
|
|
|
| 180 |
try:
|
| 181 |
text = recognizer.recognize_google(audio_data, language='fa-IR')
|
| 182 |
except sr.UnknownValueError:
|
|
|
|
| 183 |
try:
|
| 184 |
text = recognizer.recognize_google(audio_data, language='en-US')
|
| 185 |
except sr.UnknownValueError:
|
| 186 |
text = ""
|
| 187 |
except sr.RequestError:
|
| 188 |
+
text = "[خطا اتصال]"
|
| 189 |
+
if text and text != "[خطا اتصال]":
|
|
|
|
| 190 |
all_text.append(text)
|
|
|
|
| 191 |
except Exception as e:
|
| 192 |
+
print(f"File chunk error: {e}")
|
| 193 |
|
|
|
|
| 194 |
if os.path.exists(tmp_path):
|
| 195 |
os.unlink(tmp_path)
|
| 196 |
|
| 197 |
+
progress = min((i + chunk_ms) / duration_ms * 100, 100)
|
| 198 |
+
yield " ".join(all_text), f"پیشرفت: {progress:.1f}% - بخش {(i // chunk_ms)+1} از {num_chunks}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 199 |
time.sleep(0.5)
|
| 200 |
|
| 201 |
+
yield " ".join(all_text), "کامل شد! ✅"
|
|
|
|
|
|
|
| 202 |
except Exception as e:
|
| 203 |
+
yield f"خطا: {e}", "خطا ❌"
|
| 204 |
|
| 205 |
def save_text(text):
|
|
|
|
| 206 |
if not text.strip():
|
| 207 |
return None
|
|
|
|
| 208 |
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt', encoding='utf-8')
|
| 209 |
temp_file.write(text)
|
| 210 |
temp_file.close()
|
|
|
|
| 211 |
return temp_file.name
|
| 212 |
|
| 213 |
+
# Gradio UI with Timer for live updates
|
| 214 |
with gr.Blocks(
|
| 215 |
+
title="تبدیل گفتار به متن - Real-time Fixed",
|
| 216 |
theme=gr.themes.Soft(),
|
| 217 |
css="""
|
| 218 |
+
.gradio-container { font-family: 'Vazir', 'Tahoma', sans-serif !important; }
|
| 219 |
+
.rtl { direction: rtl; text-align: right; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 220 |
"""
|
| 221 |
) as demo:
|
| 222 |
gr.HTML("""
|
| 223 |
<div style="text-align: center; max-width: 800px; margin: 0 auto;">
|
| 224 |
+
<h1 style="font-size: 2.5em; margin-bottom: 0.5em;">🎤 تبدیل گفتار به متن</h1>
|
|
|
|
|
|
|
| 225 |
<p style="font-size: 1.1em; color: #666; margin-bottom: 2em;">
|
| 226 |
+
Real-time با Google API - پردازش background | قابل توزیع روی مرورگرها
|
| 227 |
</p>
|
| 228 |
</div>
|
| 229 |
""")
|
| 230 |
|
| 231 |
with gr.Tabs():
|
|
|
|
| 232 |
with gr.TabItem("🎙️ ضبط مستقیم"):
|
| 233 |
+
gr.Markdown("### فعال کنید و 5+ ثانیه واضح صحبت کنید (متن هر 3s آپدیت میشه)", elem_classes="rtl")
|
| 234 |
|
| 235 |
with gr.Row():
|
| 236 |
+
audio_input = gr.Audio(
|
| 237 |
+
sources=["microphone"],
|
| 238 |
+
type="numpy",
|
| 239 |
+
label="میکروفون (ضبط رو شروع کن)",
|
| 240 |
+
elem_classes="rtl"
|
| 241 |
+
)
|
| 242 |
+
|
| 243 |
+
start_btn = gr.Button("▶️ شروع real-time", variant="primary")
|
| 244 |
+
stop_btn = gr.Button("⏹️ توقف", variant="secondary")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 245 |
|
| 246 |
realtime_output = gr.Textbox(
|
| 247 |
+
label="متن live",
|
| 248 |
+
placeholder="پس از 3-5s صحبت، متن ظاهر میشه...",
|
| 249 |
+
lines=10,
|
| 250 |
elem_classes="rtl",
|
| 251 |
rtl=True,
|
| 252 |
+
show_copy_button=True,
|
| 253 |
+
interactive=False
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 254 |
)
|
| 255 |
|
| 256 |
+
clear_btn = gr.Button("🗑️ پاک کردن", variant="secondary")
|
|
|
|
|
|
|
|
|
|
| 257 |
|
| 258 |
+
# Events
|
| 259 |
+
audio_input.change(handle_realtime_audio, inputs=[audio_input], outputs=[audio_input]) # Handle chunks
|
|
|
|
| 260 |
|
| 261 |
+
# Timer for live update (هر 2s transcript رو pull کن)
|
| 262 |
+
timer = gr.Timer(2.0) # Start after tab open
|
| 263 |
+
timer.tick(get_current_transcript, outputs=[realtime_output])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 264 |
|
| 265 |
+
clear_btn.click(clear_transcript, outputs=[realtime_output])
|
|
|
|
|
|
|
|
|
|
|
|
|
| 266 |
|
| 267 |
+
# Start/stop recording (toggle microphone)
|
| 268 |
+
def toggle_recording(active):
|
| 269 |
+
return gr.update(value=active) # Simple toggle, but Gradio handles start/stop
|
|
|
|
|
|
|
| 270 |
|
| 271 |
+
start_btn.click(lambda: gr.update(visible=True), outputs=[stop_btn]).click(
|
| 272 |
+
toggle_recording, inputs=[audio_input], outputs=[audio_input]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 273 |
)
|
| 274 |
+
stop_btn.click(lambda: gr.update(visible=False), outputs=[start_btn])
|
| 275 |
+
|
| 276 |
+
with gr.TabItem("📁 فایل صوتی"):
|
| 277 |
+
gr.Markdown("### فایل آپلود کن و تبدیل کن", elem_classes="rtl")
|
| 278 |
|
|
|
|
| 279 |
with gr.Row():
|
| 280 |
+
file_input = gr.Audio(sources=["upload"], type="filepath", label="فایل صوتی", elem_classes="rtl")
|
| 281 |
+
chunk_slider = gr.Slider(10, 60, 30, 5, label="بخشبندی (s)", elem_classes="rtl")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 282 |
|
| 283 |
+
process_btn = gr.Button("🚀 تبدیل", variant="primary")
|
| 284 |
+
progress_label = gr.Textbox(label="وضعیت", interactive=False, elem_classes="rtl")
|
| 285 |
+
file_output = gr.Textbox(label="متن", lines=10, elem_classes="rtl", rtl=True, show_copy_button=True)
|
|
|
|
|
|
|
|
|
|
| 286 |
|
| 287 |
+
with gr.Row():
|
| 288 |
+
save_btn = gr.Button("💾 ذخیره", variant="secondary")
|
| 289 |
+
clear_file_btn = gr.Button("🗑️ پاک", variant="secondary")
|
| 290 |
+
download_file = gr.File(label="دانلود TXT", visible=False, elem_classes="rtl")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 291 |
|
| 292 |
+
process_btn.click(transcribe_file, [file_input, chunk_slider], [file_output, progress_label])
|
| 293 |
+
save_btn.click(save_text, file_output, download_file).then(lambda: gr.update(visible=True), download_file)
|
| 294 |
+
clear_file_btn.click(lambda: ("", ""), [file_output, progress_label])
|
|
|
|
| 295 |
|
| 296 |
+
with gr.Accordion("📖 راهنما", open=False, elem_classes="rtl"):
|
|
|
|
| 297 |
gr.Markdown("""
|
| 298 |
+
### استفاده:
|
| 299 |
+
- **Real-time**: میکروفون رو فعال کن، 5s+ صحبت کن. هر 3s متن آپدیت میشه (background).
|
| 300 |
+
- **فایل**: آپلود و دکمه بزن.
|
| 301 |
+
### نکات:
|
| 302 |
+
- 🗣️ واضح صحبت کن، نویز کم.
|
| 303 |
+
- 🌐 اینترنت پایدار (Google API).
|
| 304 |
+
- 📱 روی موبایل/دسکتاپ کار میکنه (mic access بده).
|
| 305 |
+
- ⚠️ محدودیت Google: ~60 req/min - برای حجم بالا، API key اضافه کن (در کد کامنت).
|
| 306 |
+
- توزیع: share لینک رو share کن، همه browserها ساپورت.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 307 |
""", elem_classes="rtl")
|
| 308 |
|
| 309 |
+
gr.HTML('<div style="text-align: center; margin-top: 2em; padding: 1em; background: #f8f9fa;"><p style="color: #666;">نسخه 2.2 - Fixed Real-time با Timer | Google Backend</p></div>')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 310 |
|
|
|
|
| 311 |
if __name__ == "__main__":
|
| 312 |
+
demo.queue().launch(share=True, show_error=True, server_name="0.0.0.0", server_port=7860)
|
|
|
|
|
|
|
|
|