Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
|
@@ -20,7 +20,7 @@ import google.generativeai as genai
|
|
| 20 |
from pydantic import BaseModel
|
| 21 |
from PIL import Image, ImageDraw, ImageFont
|
| 22 |
from pydub import AudioSegment
|
| 23 |
-
from
|
| 24 |
|
| 25 |
app = FastAPI()
|
| 26 |
|
|
@@ -61,10 +61,9 @@ FONT_FILES_MAP = {
|
|
| 61 |
"hasti": "Hasti.ttf",
|
| 62 |
}
|
| 63 |
|
| 64 |
-
#
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
NEW_SPACE_URL = os.getenv("NEW_SPACE_URL", "https://opera8-tts3.hf.space")
|
| 68 |
|
| 69 |
# --- سیستم صف و پایش موقت آپلود غیرهمگام ---
|
| 70 |
class UploadJobStatus:
|
|
@@ -111,7 +110,7 @@ def load_all_styles():
|
|
| 111 |
|
| 112 |
load_all_styles()
|
| 113 |
|
| 114 |
-
# تنظیم همزمانی
|
| 115 |
CONCURRENCY_LIMIT = 20
|
| 116 |
GEMINI_SEMAPHORE = asyncio.Semaphore(CONCURRENCY_LIMIT)
|
| 117 |
|
|
@@ -185,7 +184,6 @@ async def queue_worker():
|
|
| 185 |
job.error_message = str(e)
|
| 186 |
render_queue.task_done()
|
| 187 |
|
| 188 |
-
# کارگر پسزمینه برای پاکسازی خودکار فایلهای خروجی قدیمیتر از ۲ ساعت
|
| 189 |
async def cleanup_old_files_loop():
|
| 190 |
print("--- File Cleanup Worker Started ---")
|
| 191 |
while True:
|
|
@@ -194,11 +192,9 @@ async def cleanup_old_files_loop():
|
|
| 194 |
now = time.time()
|
| 195 |
for filename in os.listdir(TEMP_DIR):
|
| 196 |
file_path = os.path.join(TEMP_DIR, filename)
|
| 197 |
-
# شناسایی فایلهای خروجی نهایی
|
| 198 |
if filename.endswith(".mp4") and "_final_" in filename:
|
| 199 |
try:
|
| 200 |
mtime = os.path.getmtime(file_path)
|
| 201 |
-
# ۷۲۰۰ ثانیه معادل ۲ ساعت است
|
| 202 |
if now - mtime > 7200:
|
| 203 |
os.remove(file_path)
|
| 204 |
print(f"Cleanup: Removed expired final video: {filename}")
|
|
@@ -451,8 +447,6 @@ def generate_subtitle_video(data: ProcessRequest, temp_dir: str):
|
|
| 451 |
seg.words.sort(key=lambda x: x.start)
|
| 452 |
words = [w.word for w in seg.words]
|
| 453 |
|
| 454 |
-
# استفاده از منطق فریمبندی زمانی دقیق برای تمامی استایلها
|
| 455 |
-
# این کار مانع از کشیده شدن یا جابجایی زمان کلمات دیگر در صورت حذف یک کلمه میشود
|
| 456 |
SUB_FRAME_DURATION = 0.025 if data.style.name == "falling_words" else 0.05
|
| 457 |
time_cursor = start_time
|
| 458 |
ANIMATION_DURATION = 0.4
|
|
@@ -530,82 +524,31 @@ def generate_style_api(req: StylePrompt):
|
|
| 530 |
except: continue
|
| 531 |
return {"primaryColor":"#FFFFFF", "outlineColor":"#000000", "font":"vazir", "fontSize":60, "backType":"solid"}
|
| 532 |
|
| 533 |
-
|
| 534 |
-
|
| 535 |
-
|
| 536 |
-
|
| 537 |
-
- در غیر اینصورت بین ۱۵ تا ۳۰ ثانیه دنبال سکوت میگردد.
|
| 538 |
-
- اگر در این بازه سکوت مناسبی نبود، بین ۱۰ تا ۴۰ ثانیه دنبال سکوت میگردد.
|
| 539 |
-
"""
|
| 540 |
-
audio = AudioSegment.from_file(file_path)
|
| 541 |
-
length_ms = len(audio)
|
| 542 |
-
chunks_info = []
|
| 543 |
|
| 544 |
-
|
| 545 |
-
|
| 546 |
-
avg_rms = audio.rms
|
| 547 |
-
silence_threshold = avg_rms * 0.4 if avg_rms > 0 else 500
|
| 548 |
|
| 549 |
-
|
| 550 |
-
|
| 551 |
-
|
| 552 |
-
|
| 553 |
-
|
| 554 |
-
|
| 555 |
-
|
| 556 |
-
|
| 557 |
-
|
| 558 |
-
|
| 559 |
-
|
| 560 |
-
|
| 561 |
-
step = 100
|
| 562 |
|
| 563 |
-
|
| 564 |
-
|
| 565 |
-
chunk_check = window[i:i+step]
|
| 566 |
-
if chunk_check.rms < min_loudness:
|
| 567 |
-
min_loudness = chunk_check.rms
|
| 568 |
-
best_cut_rel = i
|
| 569 |
-
|
| 570 |
-
# 2. بررسی اینکه آیا سکوت پیدا شده خوب است؟
|
| 571 |
-
# و اینکه آیا اصلاً اجازه گسترش بازه تا 40 ثانیه را داریم؟ (باقیمانده > 40s)
|
| 572 |
-
if min_loudness > silence_threshold and (length_ms - start_ms) > 30 * 1000:
|
| 573 |
-
# جستجوی ثانویه در بازه 10 تا 40 ثانیه (چون سکوت خوبی پیدا نشد)
|
| 574 |
-
search_start_fb = start_ms + (10 * 1000)
|
| 575 |
-
search_end_fb = min(start_ms + (40 * 1000), length_ms)
|
| 576 |
-
window_fb = audio[search_start_fb:search_end_fb]
|
| 577 |
-
|
| 578 |
-
min_loudness_fb = float('inf')
|
| 579 |
-
best_cut_rel_fb = search_end_fb - search_start_fb
|
| 580 |
-
|
| 581 |
-
if len(window_fb) > 0:
|
| 582 |
-
for i in range(0, len(window_fb), step):
|
| 583 |
-
chunk_check = window_fb[i:i+step]
|
| 584 |
-
if chunk_check.rms < min_loudness_fb:
|
| 585 |
-
min_loudness_fb = chunk_check.rms
|
| 586 |
-
best_cut_rel_fb = i
|
| 587 |
-
|
| 588 |
-
# اعمال نتیجه جستجوی ثانویه (گستردهتر)
|
| 589 |
-
end_ms = search_start_fb + best_cut_rel_fb
|
| 590 |
-
else:
|
| 591 |
-
# اعمال نتیجه جستجوی اولیه (همان 15 تا 30 ثانیه)
|
| 592 |
-
end_ms = search_start + best_cut_rel
|
| 593 |
|
| 594 |
-
|
| 595 |
-
chunk_filename = f"{file_path}_{start_ms}.wav"
|
| 596 |
-
chunk.export(chunk_filename, format="wav", parameters=["-ar", "16000"])
|
| 597 |
-
|
| 598 |
-
chunks_info.append({
|
| 599 |
-
"path": chunk_filename,
|
| 600 |
-
"offset": start_ms / 1000.0,
|
| 601 |
-
"duration": (end_ms - start_ms) / 1000.0,
|
| 602 |
-
"index": len(chunks_info)
|
| 603 |
-
})
|
| 604 |
-
start_ms = end_ms
|
| 605 |
-
|
| 606 |
-
return chunks_info
|
| 607 |
-
|
| 608 |
-
# --- پردازش موقت در صف پسزمینه برای داک داکر غیرهمگام ---
|
| 609 |
async def bg_process_upload(task_id: str, raw_path: str, fixed_path: str, audio_path: str):
|
| 610 |
job = upload_jobs.get(task_id)
|
| 611 |
if not job:
|
|
@@ -613,6 +556,7 @@ async def bg_process_upload(task_id: str, raw_path: str, fixed_path: str, audio_
|
|
| 613 |
|
| 614 |
job.status = UploadJobStatus.PROCESSING
|
| 615 |
try:
|
|
|
|
| 616 |
proc1 = await asyncio.create_subprocess_exec(
|
| 617 |
"ffmpeg", "-y", "-i", raw_path, "-r", "30", "-c:v", "libx264",
|
| 618 |
"-preset", "ultrafast", "-c:a", "copy", fixed_path,
|
|
@@ -622,32 +566,72 @@ async def bg_process_upload(task_id: str, raw_path: str, fixed_path: str, audio_
|
|
| 622 |
|
| 623 |
w, h, total_duration = get_video_info(fixed_path)
|
| 624 |
|
|
|
|
| 625 |
proc2 = await asyncio.create_subprocess_exec(
|
| 626 |
-
"ffmpeg", "-y", "-i", fixed_path, "-vn", "-acodec", "
|
| 627 |
-
"-ar", "16000", "-ac", "1", audio_path,
|
| 628 |
stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL
|
| 629 |
)
|
| 630 |
await proc2.communicate()
|
| 631 |
|
| 632 |
-
|
| 633 |
-
audio_chunks = smart_split_audio(audio_path)
|
| 634 |
-
print(f"--- [BG Task {task_id}] Processing {len(audio_chunks)} chunks concurrently ---")
|
| 635 |
|
| 636 |
-
#
|
| 637 |
-
|
| 638 |
-
results = await asyncio.gather(*tasks)
|
| 639 |
|
| 640 |
-
|
| 641 |
-
|
| 642 |
-
|
| 643 |
-
|
| 644 |
-
|
| 645 |
-
|
| 646 |
-
|
| 647 |
-
|
| 648 |
-
|
| 649 |
-
|
| 650 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 651 |
if os.path.exists(audio_path):
|
| 652 |
try: os.remove(audio_path)
|
| 653 |
except: pass
|
|
@@ -655,19 +639,52 @@ async def bg_process_upload(task_id: str, raw_path: str, fixed_path: str, audio_
|
|
| 655 |
try: os.remove(raw_path)
|
| 656 |
except: pass
|
| 657 |
|
| 658 |
-
if not
|
| 659 |
-
err_summary = " | ".join(list(set(errors))) if errors else "هیچ متنی از تحلیل صدا استخراج نشد."
|
| 660 |
job.status = UploadJobStatus.FAILED
|
| 661 |
-
job.error_message =
|
| 662 |
return
|
| 663 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 664 |
job.result = {
|
| 665 |
"file_id": task_id,
|
| 666 |
"url": f"/temp/{task_id}.mp4",
|
| 667 |
"width": w,
|
| 668 |
"height": h,
|
| 669 |
"segments": final_segs,
|
| 670 |
-
"suggested_style":
|
| 671 |
}
|
| 672 |
job.status = UploadJobStatus.COMPLETED
|
| 673 |
|
|
@@ -681,171 +698,13 @@ async def bg_process_upload(task_id: str, raw_path: str, fixed_path: str, audio_
|
|
| 681 |
job.status = UploadJobStatus.FAILED
|
| 682 |
job.error_message = f"Processing Error: {str(e)}"
|
| 683 |
|
| 684 |
-
# --- پردازشگر ارتباط با API وبسرویس جدید در هاگینگفیس شما ---
|
| 685 |
-
async def process_single_chunk_async(chunk_data):
|
| 686 |
-
c_path = chunk_data["path"]
|
| 687 |
-
c_offset = chunk_data["offset"]
|
| 688 |
-
c_duration = chunk_data["duration"]
|
| 689 |
-
chunk_idx = chunk_data["index"]
|
| 690 |
-
|
| 691 |
-
print(f"[{chunk_idx}] Start processing chunk... Offset: {c_offset}s, Duration: {c_duration}s")
|
| 692 |
-
result_segs = []
|
| 693 |
-
style_suggestion = None
|
| 694 |
-
error_msg = None
|
| 695 |
-
|
| 696 |
-
async with GEMINI_SEMAPHORE:
|
| 697 |
-
for attempt in range(5):
|
| 698 |
-
try:
|
| 699 |
-
chunk_run_id = f"stt_chunk_{uuid.uuid4().hex[:12]}"
|
| 700 |
-
|
| 701 |
-
upload_url = f"{NEW_SPACE_URL}/api/webhook/upload"
|
| 702 |
-
print(f"[{chunk_idx}] Uploading chunk file to {upload_url} (Attempt {attempt + 1}/5)...")
|
| 703 |
-
|
| 704 |
-
with open(c_path, "rb") as f:
|
| 705 |
-
files = {"file": f}
|
| 706 |
-
data = {"run_id": chunk_run_id, "ext": "wav"}
|
| 707 |
-
upload_res = await asyncio.to_thread(
|
| 708 |
-
lambda: requests.post(upload_url, data=data, files=files, timeout=30)
|
| 709 |
-
)
|
| 710 |
-
|
| 711 |
-
if not upload_res.ok:
|
| 712 |
-
error_msg = f"خطا در بارگذاری چنک صوتی روی سرور: {upload_res.status_code}"
|
| 713 |
-
print(f"[{chunk_idx}] Warning: {error_msg}")
|
| 714 |
-
await asyncio.sleep(2)
|
| 715 |
-
continue
|
| 716 |
-
|
| 717 |
-
generate_url = f"{NEW_SPACE_URL}/api/generate"
|
| 718 |
-
file_url = f"{NEW_SPACE_URL}/static/images/{chunk_run_id}.wav"
|
| 719 |
-
|
| 720 |
-
config_prompt = f"config_url: {file_url} config_lang: Persian / فارسی config_optimize: true config_token: none"
|
| 721 |
-
|
| 722 |
-
print(f"[{chunk_idx}] Dispatching transcription action to {generate_url}...")
|
| 723 |
-
payload = {
|
| 724 |
-
"prompt": config_prompt,
|
| 725 |
-
"width": 1024,
|
| 726 |
-
"height": 1024,
|
| 727 |
-
"action_name": "matnbesada"
|
| 728 |
-
}
|
| 729 |
-
|
| 730 |
-
gen_res = await asyncio.to_thread(
|
| 731 |
-
lambda: requests.post(generate_url, json=payload, timeout=30)
|
| 732 |
-
)
|
| 733 |
-
|
| 734 |
-
if not gen_res.ok:
|
| 735 |
-
error_msg = f"خطا در ارسال درخواست رندر به سرور: {gen_res.status_code}"
|
| 736 |
-
print(f"[{chunk_idx}] Warning: {error_msg}")
|
| 737 |
-
await asyncio.sleep(2)
|
| 738 |
-
continue
|
| 739 |
-
|
| 740 |
-
gen_data = gen_res.json()
|
| 741 |
-
if gen_data.get("status") != "success" or not gen_data.get("run_id"):
|
| 742 |
-
error_msg = "پاسخ نامعتبر در ثبت تسک پردازش."
|
| 743 |
-
print(f"[{chunk_idx}] Warning: {error_msg}")
|
| 744 |
-
await asyncio.sleep(2)
|
| 745 |
-
continue
|
| 746 |
-
|
| 747 |
-
action_run_id = gen_data["run_id"]
|
| 748 |
-
print(f"[{chunk_idx}] Action dispatched successfully. Action ID: {action_run_id}")
|
| 749 |
-
|
| 750 |
-
json_url = f"{NEW_SPACE_URL}/static/images/{action_run_id}.json"
|
| 751 |
-
|
| 752 |
-
print(f"[{chunk_idx}] Starting status polling for json file...")
|
| 753 |
-
is_ready = False
|
| 754 |
-
poll_attempts = 0
|
| 755 |
-
max_poll_attempts = 100
|
| 756 |
-
|
| 757 |
-
while poll_attempts < max_poll_attempts:
|
| 758 |
-
poll_attempts += 1
|
| 759 |
-
check_res = await asyncio.to_thread(
|
| 760 |
-
lambda: requests.head(json_url, timeout=10)
|
| 761 |
-
)
|
| 762 |
-
if check_res.ok:
|
| 763 |
-
is_ready = True
|
| 764 |
-
break
|
| 765 |
-
await asyncio.sleep(3)
|
| 766 |
-
|
| 767 |
-
if not is_ready:
|
| 768 |
-
error_msg = "زمان انتظار برای دریافت خروجی به پایان رسید."
|
| 769 |
-
print(f"[{chunk_idx}] Warning: {error_msg}")
|
| 770 |
-
continue
|
| 771 |
-
|
| 772 |
-
print(f"[{chunk_idx}] Files are ready! Fetching results...")
|
| 773 |
-
fetch_json_res = await asyncio.to_thread(lambda: requests.get(json_url, timeout=15))
|
| 774 |
-
|
| 775 |
-
raw_word_items = []
|
| 776 |
-
if fetch_json_res.ok:
|
| 777 |
-
try:
|
| 778 |
-
words_data = json.loads(fetch_json_res.text)
|
| 779 |
-
if isinstance(words_data, dict):
|
| 780 |
-
if "words" in words_data:
|
| 781 |
-
words_data = words_data["words"]
|
| 782 |
-
elif "segments" in words_data:
|
| 783 |
-
words_data = words_data["segments"]
|
| 784 |
-
else:
|
| 785 |
-
for val in words_data.values():
|
| 786 |
-
if isinstance(val, list):
|
| 787 |
-
words_data = val
|
| 788 |
-
break
|
| 789 |
-
if isinstance(words_data, list):
|
| 790 |
-
raw_word_items = words_data
|
| 791 |
-
except Exception as je:
|
| 792 |
-
print(f"[{chunk_idx}] JSON decode failed: {je}")
|
| 793 |
-
|
| 794 |
-
if raw_word_items:
|
| 795 |
-
print(f"[{chunk_idx}] Transcribed successfully. Processing {len(raw_word_items)} raw words...")
|
| 796 |
-
|
| 797 |
-
chunk_size = 7
|
| 798 |
-
for k in range(0, len(raw_word_items), chunk_size):
|
| 799 |
-
sub = raw_word_items[k:k+chunk_size]
|
| 800 |
-
if not sub: continue
|
| 801 |
-
sc_words = []
|
| 802 |
-
for w in sub:
|
| 803 |
-
if not isinstance(w, dict): continue
|
| 804 |
-
|
| 805 |
-
word_text = w.get("word", w.get("text", ""))
|
| 806 |
-
word_start = float(w.get("start", w.get("start_time", 0)))
|
| 807 |
-
word_end = float(w.get("end", w.get("end_time", 0)))
|
| 808 |
-
|
| 809 |
-
sc_words.append({
|
| 810 |
-
"word": word_text,
|
| 811 |
-
"start": round(c_offset + word_start, 3),
|
| 812 |
-
"end": round(c_offset + word_end, 3),
|
| 813 |
-
"highlight": False
|
| 814 |
-
})
|
| 815 |
-
|
| 816 |
-
if sc_words:
|
| 817 |
-
result_segs.append({
|
| 818 |
-
"start": sc_words[0]["start"],
|
| 819 |
-
"end": sc_words[-1]["end"],
|
| 820 |
-
"text": " ".join([x["word"] for x in sc_words]),
|
| 821 |
-
"words": sc_words
|
| 822 |
-
})
|
| 823 |
-
|
| 824 |
-
print(f"[{chunk_idx}] Successfully generated {len(result_segs)} subtitle segments.")
|
| 825 |
-
error_msg = None
|
| 826 |
-
break
|
| 827 |
-
else:
|
| 828 |
-
error_msg = "هیچ زمانبندی خامی از سرور دریافت نشد."
|
| 829 |
-
print(f"[{chunk_idx}] Warning: {error_msg}")
|
| 830 |
-
|
| 831 |
-
except Exception as e:
|
| 832 |
-
error_msg = f"{type(e).__name__}: {str(e)}"
|
| 833 |
-
print(f"[{chunk_idx}] ERROR calling custom Space STT: {error_msg}")
|
| 834 |
-
await asyncio.sleep(2 + attempt * 2)
|
| 835 |
-
|
| 836 |
-
if os.path.exists(c_path):
|
| 837 |
-
try: os.remove(c_path)
|
| 838 |
-
except: pass
|
| 839 |
-
|
| 840 |
-
return result_segs, style_suggestion, error_msg
|
| 841 |
-
|
| 842 |
@app.post("/api/upload")
|
| 843 |
async def upload(background_tasks: BackgroundTasks, file: UploadFile = File(...)):
|
| 844 |
task_id = str(uuid.uuid4())[:8]
|
| 845 |
ext = file.filename.split('.')[-1]
|
| 846 |
raw_path = f"{TEMP_DIR}/{task_id}_raw.{ext}"
|
| 847 |
fixed_path = f"{TEMP_DIR}/{task_id}.mp4"
|
| 848 |
-
audio_path = f"{TEMP_DIR}/{task_id}.
|
| 849 |
|
| 850 |
try:
|
| 851 |
with open(raw_path, "wb") as f:
|
|
@@ -884,13 +743,11 @@ async def delete_project_files(file_id: str):
|
|
| 884 |
if "/" in file_id or "\\" in file_id or ".." in file_id:
|
| 885 |
raise HTTPException(400, "Invalid file_id")
|
| 886 |
|
| 887 |
-
# ۱. حذف ویدیوی اصلی آپلود شده
|
| 888 |
raw_video_path = os.path.join(TEMP_DIR, f"{file_id}.mp4")
|
| 889 |
if os.path.exists(raw_video_path):
|
| 890 |
try: os.remove(raw_video_path)
|
| 891 |
except Exception as e: print(f"Error removing raw video: {e}")
|
| 892 |
|
| 893 |
-
# ۲. حذف تمامی خروجیهای رندر شده مرتبط با این پروژه
|
| 894 |
final_patterns = os.path.join(TEMP_DIR, f"{file_id}_final_*.mp4")
|
| 895 |
for f_path in glob.glob(final_patterns):
|
| 896 |
try: os.remove(f_path)
|
|
|
|
| 20 |
from pydantic import BaseModel
|
| 21 |
from PIL import Image, ImageDraw, ImageFont
|
| 22 |
from pydub import AudioSegment
|
| 23 |
+
from groq import Groq # اضافه شدن کتابخانه Groq
|
| 24 |
|
| 25 |
app = FastAPI()
|
| 26 |
|
|
|
|
| 61 |
"hasti": "Hasti.ttf",
|
| 62 |
}
|
| 63 |
|
| 64 |
+
# دریافت توکن Groq و Gemini از تنظیمات سکرت اسپیس
|
| 65 |
+
GROQ_API_KEY = os.getenv("GROQ_API_KEY", "")
|
| 66 |
+
API_KEYS = [os.getenv("GEMINI_API_KEY")] if os.getenv("GEMINI_API_KEY") else []
|
|
|
|
| 67 |
|
| 68 |
# --- سیستم صف و پایش موقت آپلود غیرهمگام ---
|
| 69 |
class UploadJobStatus:
|
|
|
|
| 110 |
|
| 111 |
load_all_styles()
|
| 112 |
|
| 113 |
+
# تنظیم همزمانی پردازش
|
| 114 |
CONCURRENCY_LIMIT = 20
|
| 115 |
GEMINI_SEMAPHORE = asyncio.Semaphore(CONCURRENCY_LIMIT)
|
| 116 |
|
|
|
|
| 184 |
job.error_message = str(e)
|
| 185 |
render_queue.task_done()
|
| 186 |
|
|
|
|
| 187 |
async def cleanup_old_files_loop():
|
| 188 |
print("--- File Cleanup Worker Started ---")
|
| 189 |
while True:
|
|
|
|
| 192 |
now = time.time()
|
| 193 |
for filename in os.listdir(TEMP_DIR):
|
| 194 |
file_path = os.path.join(TEMP_DIR, filename)
|
|
|
|
| 195 |
if filename.endswith(".mp4") and "_final_" in filename:
|
| 196 |
try:
|
| 197 |
mtime = os.path.getmtime(file_path)
|
|
|
|
| 198 |
if now - mtime > 7200:
|
| 199 |
os.remove(file_path)
|
| 200 |
print(f"Cleanup: Removed expired final video: {filename}")
|
|
|
|
| 447 |
seg.words.sort(key=lambda x: x.start)
|
| 448 |
words = [w.word for w in seg.words]
|
| 449 |
|
|
|
|
|
|
|
| 450 |
SUB_FRAME_DURATION = 0.025 if data.style.name == "falling_words" else 0.05
|
| 451 |
time_cursor = start_time
|
| 452 |
ANIMATION_DURATION = 0.4
|
|
|
|
| 524 |
except: continue
|
| 525 |
return {"primaryColor":"#FFFFFF", "outlineColor":"#000000", "font":"vazir", "fontSize":60, "backType":"solid"}
|
| 526 |
|
| 527 |
+
# --- تابع جدید برای ارسال فایل صوتی به Groq و تحلیل دقیق کلمهبهکلمه ---
|
| 528 |
+
async def transcribe_audio_via_groq(audio_path: str):
|
| 529 |
+
if not GROQ_API_KEY:
|
| 530 |
+
raise Exception("متغیر محیطی GROQ_API_KEY در قسمت Secrets تنظیم نشده است.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 531 |
|
| 532 |
+
# راهاندازی کلاینت Groq
|
| 533 |
+
client = Groq(api_key=GROQ_API_KEY)
|
|
|
|
|
|
|
| 534 |
|
| 535 |
+
# فراخوانی متد همگام به صورت غ��رهمگام با استفاده از نخ مستقل
|
| 536 |
+
def call_groq():
|
| 537 |
+
with open(audio_path, "rb") as file:
|
| 538 |
+
return client.audio.transcriptions.create(
|
| 539 |
+
file=(os.path.basename(audio_path), file.read()),
|
| 540 |
+
model="whisper-large-v3",
|
| 541 |
+
response_format="verbose_json",
|
| 542 |
+
timestamp_granularities=["word"],
|
| 543 |
+
temperature=0.0,
|
| 544 |
+
language="fa", # تضمین پردازش به عنوان زبان فارسی
|
| 545 |
+
prompt="دقیقترین زیرنویس چسبیده فارسی با رعایت علائم نگارشی، فاصلهگذاری، نیمفاصلهها و بدون جا انداختن کلمات."
|
| 546 |
+
)
|
|
|
|
| 547 |
|
| 548 |
+
transcription = await asyncio.to_thread(call_groq)
|
| 549 |
+
return transcription
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 550 |
|
| 551 |
+
# --- پردازش موقت در صف پسزمینه ---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 552 |
async def bg_process_upload(task_id: str, raw_path: str, fixed_path: str, audio_path: str):
|
| 553 |
job = upload_jobs.get(task_id)
|
| 554 |
if not job:
|
|
|
|
| 556 |
|
| 557 |
job.status = UploadJobStatus.PROCESSING
|
| 558 |
try:
|
| 559 |
+
# ۱. تبدیل ویدیوی خام به فرمت استاندارد 30fps h264
|
| 560 |
proc1 = await asyncio.create_subprocess_exec(
|
| 561 |
"ffmpeg", "-y", "-i", raw_path, "-r", "30", "-c:v", "libx264",
|
| 562 |
"-preset", "ultrafast", "-c:a", "copy", fixed_path,
|
|
|
|
| 566 |
|
| 567 |
w, h, total_duration = get_video_info(fixed_path)
|
| 568 |
|
| 569 |
+
# ۲. استخراج مستقیم صدا به صورت MP3 با نرخ بیت پایینتر (کیفیت عالی برای Whisper و حجم بسیار کم زیر ۲۵ مگابایت)
|
| 570 |
proc2 = await asyncio.create_subprocess_exec(
|
| 571 |
+
"ffmpeg", "-y", "-i", fixed_path, "-vn", "-acodec", "libmp3lame",
|
| 572 |
+
"-ar", "16000", "-ac", "1", "-b:a", "64k", audio_path,
|
| 573 |
stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL
|
| 574 |
)
|
| 575 |
await proc2.communicate()
|
| 576 |
|
| 577 |
+
print(f"--- [BG Task {task_id}] Transcribing audio with Groq Whisper-large-v3 in one step ---")
|
|
|
|
|
|
|
| 578 |
|
| 579 |
+
# ۳. ارسال مستقیم فایل صوتی فشرده شده به Groq
|
| 580 |
+
transcription = await transcribe_audio_via_groq(audio_path)
|
|
|
|
| 581 |
|
| 582 |
+
# ۴. تحلیل خروجی Pydantic یا دیکشنری دریافتی
|
| 583 |
+
if hasattr(transcription, "model_dump"):
|
| 584 |
+
data = transcription.model_dump()
|
| 585 |
+
elif hasattr(transcription, "dict"):
|
| 586 |
+
data = transcription.dict()
|
| 587 |
+
elif isinstance(transcription, dict):
|
| 588 |
+
data = transcription
|
| 589 |
+
else:
|
| 590 |
+
try:
|
| 591 |
+
data = dict(transcription)
|
| 592 |
+
except:
|
| 593 |
+
data = {}
|
| 594 |
+
if hasattr(transcription, "words"):
|
| 595 |
+
data["words"] = getattr(transcription, "words", [])
|
| 596 |
+
if hasattr(transcription, "segments"):
|
| 597 |
+
data["segments"] = getattr(transcription, "segments", [])
|
| 598 |
+
|
| 599 |
+
raw_word_items = data.get("words", [])
|
| 600 |
+
# در صورت نبود آرایه مستقیم کلمات، آن را از میان سگمنتها استخراج میکنیم
|
| 601 |
+
if not raw_word_items and "segments" in data:
|
| 602 |
+
segments = data.get("segments", [])
|
| 603 |
+
for seg in segments:
|
| 604 |
+
if isinstance(seg, dict) and "words" in seg:
|
| 605 |
+
raw_word_items.extend(seg["words"])
|
| 606 |
+
elif hasattr(seg, "words"):
|
| 607 |
+
raw_word_items.extend(getattr(seg, "words", []))
|
| 608 |
+
|
| 609 |
+
# حالت بکآپ (اگر کلمهبهکلمه خالی بود، از سطح سگمنت کلمات را استخراج میکنیم)
|
| 610 |
+
if not raw_word_items:
|
| 611 |
+
segments = data.get("segments", [])
|
| 612 |
+
if segments:
|
| 613 |
+
print(f"[{task_id}] No word timestamps found. Falling back to segment level timestamps.")
|
| 614 |
+
for seg in segments:
|
| 615 |
+
seg_text = getattr(seg, "text", "") if not isinstance(seg, dict) else seg.get("text", "")
|
| 616 |
+
seg_start = float(getattr(seg, "start", 0)) if not isinstance(seg, dict) else float(seg.get("start", 0))
|
| 617 |
+
seg_end = float(getattr(seg, "end", 0)) if not isinstance(seg, dict) else float(seg.get("end", 0))
|
| 618 |
+
words_list = seg_text.split()
|
| 619 |
+
duration = seg_end - seg_start
|
| 620 |
+
word_duration = duration / max(1, len(words_list))
|
| 621 |
+
|
| 622 |
+
sc_words = []
|
| 623 |
+
for i, w in enumerate(words_list):
|
| 624 |
+
w_start = seg_start + (i * word_duration)
|
| 625 |
+
w_end = w_start + word_duration
|
| 626 |
+
sc_words.append({
|
| 627 |
+
"word": w,
|
| 628 |
+
"start": round(w_start, 3),
|
| 629 |
+
"end": round(w_end, 3),
|
| 630 |
+
"highlight": False
|
| 631 |
+
})
|
| 632 |
+
raw_word_items.extend(sc_words)
|
| 633 |
+
|
| 634 |
+
# پاکسازی سریع فایلهای موقت صوتی و ویدئوی خام
|
| 635 |
if os.path.exists(audio_path):
|
| 636 |
try: os.remove(audio_path)
|
| 637 |
except: pass
|
|
|
|
| 639 |
try: os.remove(raw_path)
|
| 640 |
except: pass
|
| 641 |
|
| 642 |
+
if not raw_word_items:
|
|
|
|
| 643 |
job.status = UploadJobStatus.FAILED
|
| 644 |
+
job.error_message = "تبدیل گفتار به متن ناموفق بود. هیچ متنی شناسایی نشد."
|
| 645 |
return
|
| 646 |
|
| 647 |
+
# ۵. گروهبندی کلمات دریافت شده به صورت دستههای ۵ تایی (بسیار خوانا و استاندارد برای ریلز)
|
| 648 |
+
final_segs = []
|
| 649 |
+
chunk_size = 5
|
| 650 |
+
for k in range(0, len(raw_word_items), chunk_size):
|
| 651 |
+
sub = raw_word_items[k:k+chunk_size]
|
| 652 |
+
if not sub: continue
|
| 653 |
+
sc_words = []
|
| 654 |
+
for w in sub:
|
| 655 |
+
if not isinstance(w, dict):
|
| 656 |
+
word_text = getattr(w, "word", getattr(w, "text", ""))
|
| 657 |
+
word_start = float(getattr(w, "start", getattr(w, "start_time", 0)))
|
| 658 |
+
word_end = float(getattr(w, "end", getattr(w, "end_time", 0)))
|
| 659 |
+
else:
|
| 660 |
+
word_text = w.get("word", w.get("text", ""))
|
| 661 |
+
word_start = float(w.get("start", w.get("start_time", 0)))
|
| 662 |
+
word_end = float(w.get("end", w.get("end_time", 0)))
|
| 663 |
+
|
| 664 |
+
sc_words.append({
|
| 665 |
+
"word": word_text,
|
| 666 |
+
"start": round(word_start, 3),
|
| 667 |
+
"end": round(word_end, 3),
|
| 668 |
+
"highlight": False
|
| 669 |
+
})
|
| 670 |
+
|
| 671 |
+
if sc_words:
|
| 672 |
+
final_segs.append({
|
| 673 |
+
"start": sc_words[0]["start"],
|
| 674 |
+
"end": sc_words[-1]["end"],
|
| 675 |
+
"text": " ".join([x["word"] for x in sc_words]),
|
| 676 |
+
"words": sc_words
|
| 677 |
+
})
|
| 678 |
+
|
| 679 |
+
suggested_style = {"primaryColor": "#FFFFFF", "outlineColor": "#000000", "font": "vazir", "fontSize": 60, "backType": "solid"}
|
| 680 |
+
|
| 681 |
job.result = {
|
| 682 |
"file_id": task_id,
|
| 683 |
"url": f"/temp/{task_id}.mp4",
|
| 684 |
"width": w,
|
| 685 |
"height": h,
|
| 686 |
"segments": final_segs,
|
| 687 |
+
"suggested_style": suggested_style
|
| 688 |
}
|
| 689 |
job.status = UploadJobStatus.COMPLETED
|
| 690 |
|
|
|
|
| 698 |
job.status = UploadJobStatus.FAILED
|
| 699 |
job.error_message = f"Processing Error: {str(e)}"
|
| 700 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 701 |
@app.post("/api/upload")
|
| 702 |
async def upload(background_tasks: BackgroundTasks, file: UploadFile = File(...)):
|
| 703 |
task_id = str(uuid.uuid4())[:8]
|
| 704 |
ext = file.filename.split('.')[-1]
|
| 705 |
raw_path = f"{TEMP_DIR}/{task_id}_raw.{ext}"
|
| 706 |
fixed_path = f"{TEMP_DIR}/{task_id}.mp4"
|
| 707 |
+
audio_path = f"{TEMP_DIR}/{task_id}.mp3" # ذخیره به عنوان MP3
|
| 708 |
|
| 709 |
try:
|
| 710 |
with open(raw_path, "wb") as f:
|
|
|
|
| 743 |
if "/" in file_id or "\\" in file_id or ".." in file_id:
|
| 744 |
raise HTTPException(400, "Invalid file_id")
|
| 745 |
|
|
|
|
| 746 |
raw_video_path = os.path.join(TEMP_DIR, f"{file_id}.mp4")
|
| 747 |
if os.path.exists(raw_video_path):
|
| 748 |
try: os.remove(raw_video_path)
|
| 749 |
except Exception as e: print(f"Error removing raw video: {e}")
|
| 750 |
|
|
|
|
| 751 |
final_patterns = os.path.join(TEMP_DIR, f"{file_id}_final_*.mp4")
|
| 752 |
for f_path in glob.glob(final_patterns):
|
| 753 |
try: os.remove(f_path)
|