Spaces:
Running
Running
simple version
Browse files
app.py
CHANGED
|
@@ -1,117 +1,50 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
-
import cv2
|
| 3 |
-
import mediapipe as mp
|
| 4 |
-
import whisper
|
| 5 |
-
import numpy as np
|
| 6 |
import subprocess
|
| 7 |
-
import os
|
| 8 |
-
from collections import deque
|
| 9 |
import tempfile
|
| 10 |
-
import
|
| 11 |
-
|
| 12 |
-
# -----------------------------
|
| 13 |
-
# تنظیم MediaPipe
|
| 14 |
-
# -----------------------------
|
| 15 |
-
mp_face_mesh = mp.solutions.face_mesh
|
| 16 |
-
face_mesh = mp_face_mesh.FaceMesh(static_image_mode=False, max_num_faces=1)
|
| 17 |
-
|
| 18 |
-
# -----------------------------
|
| 19 |
-
# بارگذاری مدلهای Haar Cascade
|
| 20 |
-
# -----------------------------
|
| 21 |
-
face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + "haarcascade_frontalface_default.xml")
|
| 22 |
-
smile_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + "haarcascade_smile.xml")
|
| 23 |
-
|
| 24 |
-
# -----------------------------
|
| 25 |
-
# بارگذاری مدل Whisper Large
|
| 26 |
-
# -----------------------------
|
| 27 |
-
print("🎧 Loading Whisper model (large)...")
|
| 28 |
-
model = whisper.load_model("large")
|
| 29 |
|
| 30 |
# -----------------------------
|
| 31 |
-
#
|
| 32 |
# -----------------------------
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
bottom_lip = landmarks[14]
|
| 36 |
-
left_corner = landmarks[61]
|
| 37 |
-
right_corner = landmarks[291]
|
| 38 |
-
return (bottom_lip.y - top_lip.y) / (right_corner.x - left_corner.x)
|
| 39 |
|
| 40 |
# -----------------------------
|
| 41 |
-
# تابع پردازش
|
| 42 |
# -----------------------------
|
| 43 |
-
def
|
| 44 |
-
#
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
else:
|
| 48 |
-
tmp_video = tempfile.NamedTemporaryFile(delete=False, suffix=".mp4")
|
| 49 |
-
tmp_video.write(video_path.read())
|
| 50 |
-
tmp_video.close()
|
| 51 |
tmp_video_path = tmp_video.name
|
| 52 |
|
| 53 |
-
# استخراج صدا
|
| 54 |
-
|
| 55 |
-
subprocess.
|
| 56 |
-
[
|
| 57 |
stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL
|
| 58 |
)
|
| 59 |
|
| 60 |
-
#
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
cap = cv2.VideoCapture(tmp_video_path)
|
| 65 |
-
har_history = deque(maxlen=5)
|
| 66 |
-
audio_index = 0
|
| 67 |
-
transcribed_text = ""
|
| 68 |
-
chunk_size = int(sr * 2) # هر 2 ثانیه صدا
|
| 69 |
-
|
| 70 |
-
while cap.isOpened():
|
| 71 |
-
ret, frame = cap.read()
|
| 72 |
-
if not ret:
|
| 73 |
-
break
|
| 74 |
-
|
| 75 |
-
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
|
| 76 |
-
faces = face_cascade.detectMultiScale(gray, 1.3, 5)
|
| 77 |
-
speaking_flag = False
|
| 78 |
-
|
| 79 |
-
for (x, y, w, h) in faces:
|
| 80 |
-
roi_rgb = cv2.cvtColor(frame[y:y+h, x:x+w], cv2.COLOR_BGR2RGB)
|
| 81 |
-
results = face_mesh.process(roi_rgb)
|
| 82 |
-
smiles = smile_cascade.detectMultiScale(gray[y:y+h, x:x+w], scaleFactor=1.7, minNeighbors=20)
|
| 83 |
-
|
| 84 |
-
if results.multi_face_landmarks:
|
| 85 |
-
for face_landmarks in results.multi_face_landmarks:
|
| 86 |
-
landmarks = face_landmarks.landmark
|
| 87 |
-
har = lip_aspect_ratio(landmarks, w, h)
|
| 88 |
-
har_history.append(har)
|
| 89 |
-
avg_har = np.mean(har_history)
|
| 90 |
-
speaking_flag = avg_har > 0.3 or len(smiles) > 0
|
| 91 |
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
if len(chunk) > 0:
|
| 96 |
-
# Whisper expects a file, so save chunk temporarily
|
| 97 |
-
tmp_chunk_path = tempfile.NamedTemporaryFile(delete=False, suffix=".wav").name
|
| 98 |
-
sf.write(tmp_chunk_path, chunk, sr)
|
| 99 |
-
result = model.transcribe(tmp_chunk_path, fp16=False)
|
| 100 |
-
transcribed_text += result["text"] + " "
|
| 101 |
-
audio_index += chunk_size
|
| 102 |
|
| 103 |
-
|
| 104 |
-
return transcribed_text.strip()
|
| 105 |
|
| 106 |
# -----------------------------
|
| 107 |
-
# رابط Gradio
|
| 108 |
# -----------------------------
|
| 109 |
iface = gr.Interface(
|
| 110 |
-
fn=
|
| 111 |
-
inputs=gr.Video(label="
|
| 112 |
-
outputs=gr.Textbox(label="
|
| 113 |
-
title="
|
| 114 |
-
description="
|
| 115 |
)
|
| 116 |
|
| 117 |
if __name__ == "__main__":
|
|
|
|
| 1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
import subprocess
|
|
|
|
|
|
|
| 3 |
import tempfile
|
| 4 |
+
import whisper
|
| 5 |
+
import os
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
|
| 7 |
# -----------------------------
|
| 8 |
+
# بارگذاری مدل Whisper
|
| 9 |
# -----------------------------
|
| 10 |
+
print("🎧 Loading Whisper model (base)...")
|
| 11 |
+
model = whisper.load_model("base") # میتوانید "small", "medium", یا "large" هم بگذارید
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
# -----------------------------
|
| 14 |
+
# تابع پردازش ویدیو و تبدیل به متن
|
| 15 |
# -----------------------------
|
| 16 |
+
def transcribe_video(video):
|
| 17 |
+
# ذخیره موقت ویدیو
|
| 18 |
+
with tempfile.NamedTemporaryFile(delete=False, suffix=".mp4") as tmp_video:
|
| 19 |
+
tmp_video.write(video.read())
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
tmp_video_path = tmp_video.name
|
| 21 |
|
| 22 |
+
# استخراج صدا با ffmpeg
|
| 23 |
+
tmp_audio_path = tempfile.NamedTemporaryFile(delete=False, suffix=".wav").name
|
| 24 |
+
subprocess.run(
|
| 25 |
+
["ffmpeg", "-y", "-i", tmp_video_path, "-ar", "16000", "-ac", "1", tmp_audio_path],
|
| 26 |
stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL
|
| 27 |
)
|
| 28 |
|
| 29 |
+
# تبدیل صدا به متن
|
| 30 |
+
result = model.transcribe(tmp_audio_path, fp16=False)
|
| 31 |
+
text = result["text"].strip()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
|
| 33 |
+
# پاک کردن فایلهای موقت
|
| 34 |
+
os.remove(tmp_video_path)
|
| 35 |
+
os.remove(tmp_audio_path)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
|
| 37 |
+
return text or "متنی شناسایی نشد."
|
|
|
|
| 38 |
|
| 39 |
# -----------------------------
|
| 40 |
+
# رابط کاربری Gradio
|
| 41 |
# -----------------------------
|
| 42 |
iface = gr.Interface(
|
| 43 |
+
fn=transcribe_video,
|
| 44 |
+
inputs=gr.Video(label="🎥 ویدیو را آپلود کنید"),
|
| 45 |
+
outputs=gr.Textbox(label="📝 متن استخراجشده"),
|
| 46 |
+
title="🎧 ویدیو به متن با Whisper",
|
| 47 |
+
description="ویدیوی خود را آپلود کنید تا گفتار آن به متن تبدیل شود."
|
| 48 |
)
|
| 49 |
|
| 50 |
if __name__ == "__main__":
|