Upload app.py with huggingface_hub
Browse files
app.py
ADDED
|
@@ -0,0 +1,501 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import torch
|
| 3 |
+
import torchaudio
|
| 4 |
+
import whisper
|
| 5 |
+
import cv2
|
| 6 |
+
import numpy as np
|
| 7 |
+
from moviepy.editor import VideoFileClip, TextClip, CompositeVideoClip
|
| 8 |
+
from transformers import pipeline, AutoTokenizer, AutoModel
|
| 9 |
+
import tempfile
|
| 10 |
+
import os
|
| 11 |
+
import json
|
| 12 |
+
from datetime import timedelta
|
| 13 |
+
import librosa
|
| 14 |
+
from scipy.signal import find_peaks
|
| 15 |
+
import tensorflow as tf
|
| 16 |
+
from sklearn.feature_extraction.text import TfidfVectorizer
|
| 17 |
+
from sklearn.metrics.pairwise import cosine_similarity
|
| 18 |
+
import spacy
|
| 19 |
+
import nltk
|
| 20 |
+
from googletrans import Translator
|
| 21 |
+
import warnings
|
| 22 |
+
warnings.filterwarnings("ignore")
|
| 23 |
+
|
| 24 |
+
class ZenVisionModel:
|
| 25 |
+
"""
|
| 26 |
+
ZenVision - Advanced AI Subtitle Generation Model
|
| 27 |
+
Desarrollado por el equipo ZenVision
|
| 28 |
+
Modelo de 3GB+ con múltiples tecnologías de IA
|
| 29 |
+
"""
|
| 30 |
+
|
| 31 |
+
def __init__(self):
|
| 32 |
+
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 33 |
+
print(f"🚀 Inicializando ZenVision en {self.device}")
|
| 34 |
+
|
| 35 |
+
# Cargar modelos de IA
|
| 36 |
+
self.load_models()
|
| 37 |
+
|
| 38 |
+
def load_models(self):
|
| 39 |
+
"""Carga todos los modelos de IA necesarios"""
|
| 40 |
+
print("📦 Cargando modelos de IA...")
|
| 41 |
+
|
| 42 |
+
# 1. Whisper para transcripción de audio (1.5GB)
|
| 43 |
+
self.whisper_model = whisper.load_model("large-v2")
|
| 44 |
+
|
| 45 |
+
# 2. Modelo de traducción multiidioma (500MB)
|
| 46 |
+
self.translator = pipeline("translation",
|
| 47 |
+
model="Helsinki-NLP/opus-mt-en-mul",
|
| 48 |
+
device=0 if self.device == "cuda" else -1)
|
| 49 |
+
|
| 50 |
+
# 3. Modelo de análisis de sentimientos (200MB)
|
| 51 |
+
self.sentiment_analyzer = pipeline("sentiment-analysis",
|
| 52 |
+
model="cardiffnlp/twitter-roberta-base-sentiment-latest",
|
| 53 |
+
device=0 if self.device == "cuda" else -1)
|
| 54 |
+
|
| 55 |
+
# 4. Modelo de detección de emociones (300MB)
|
| 56 |
+
self.emotion_detector = pipeline("text-classification",
|
| 57 |
+
model="j-hartmann/emotion-english-distilroberta-base",
|
| 58 |
+
device=0 if self.device == "cuda" else -1)
|
| 59 |
+
|
| 60 |
+
# 5. Modelo BERT para embeddings (400MB)
|
| 61 |
+
self.bert_tokenizer = AutoTokenizer.from_pretrained("bert-base-multilingual-cased")
|
| 62 |
+
self.bert_model = AutoModel.from_pretrained("bert-base-multilingual-cased")
|
| 63 |
+
|
| 64 |
+
# 6. Traductor de Google
|
| 65 |
+
self.google_translator = Translator()
|
| 66 |
+
|
| 67 |
+
# 7. Procesador de lenguaje natural
|
| 68 |
+
try:
|
| 69 |
+
self.nlp = spacy.load("en_core_web_sm")
|
| 70 |
+
except:
|
| 71 |
+
print("⚠️ Modelo spacy no encontrado, usando funcionalidad básica")
|
| 72 |
+
self.nlp = None
|
| 73 |
+
|
| 74 |
+
print("✅ Todos los modelos cargados exitosamente")
|
| 75 |
+
|
| 76 |
+
def extract_audio_features(self, video_path):
|
| 77 |
+
"""Extrae características avanzadas del audio"""
|
| 78 |
+
print("🎵 Extrayendo características de audio...")
|
| 79 |
+
|
| 80 |
+
# Extraer audio del video
|
| 81 |
+
video = VideoFileClip(video_path)
|
| 82 |
+
audio_path = tempfile.mktemp(suffix=".wav")
|
| 83 |
+
video.audio.write_audiofile(audio_path, verbose=False, logger=None)
|
| 84 |
+
|
| 85 |
+
# Cargar audio con librosa para análisis avanzado
|
| 86 |
+
y, sr = librosa.load(audio_path, sr=16000)
|
| 87 |
+
|
| 88 |
+
# Características espectrales
|
| 89 |
+
mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13)
|
| 90 |
+
spectral_centroids = librosa.feature.spectral_centroid(y=y, sr=sr)
|
| 91 |
+
chroma = librosa.feature.chroma_stft(y=y, sr=sr)
|
| 92 |
+
|
| 93 |
+
# Detección de pausas y segmentos
|
| 94 |
+
intervals = librosa.effects.split(y, top_db=20)
|
| 95 |
+
|
| 96 |
+
video.close()
|
| 97 |
+
os.remove(audio_path)
|
| 98 |
+
|
| 99 |
+
return {
|
| 100 |
+
'audio_data': y,
|
| 101 |
+
'sample_rate': sr,
|
| 102 |
+
'mfccs': mfccs,
|
| 103 |
+
'spectral_centroids': spectral_centroids,
|
| 104 |
+
'chroma': chroma,
|
| 105 |
+
'intervals': intervals,
|
| 106 |
+
'duration': len(y) / sr
|
| 107 |
+
}
|
| 108 |
+
|
| 109 |
+
def advanced_transcription(self, audio_features):
|
| 110 |
+
"""Transcripción avanzada con Whisper y análisis contextual"""
|
| 111 |
+
print("🎤 Realizando transcripción avanzada...")
|
| 112 |
+
|
| 113 |
+
# Transcripción con Whisper
|
| 114 |
+
result = self.whisper_model.transcribe(
|
| 115 |
+
audio_features['audio_data'],
|
| 116 |
+
language="auto",
|
| 117 |
+
word_timestamps=True,
|
| 118 |
+
verbose=False
|
| 119 |
+
)
|
| 120 |
+
|
| 121 |
+
# Procesar segmentos con timestamps precisos
|
| 122 |
+
segments = []
|
| 123 |
+
for segment in result['segments']:
|
| 124 |
+
# Análisis de sentimientos del texto
|
| 125 |
+
sentiment = self.sentiment_analyzer(segment['text'])[0]
|
| 126 |
+
|
| 127 |
+
# Análisis de emociones
|
| 128 |
+
emotion = self.emotion_detector(segment['text'])[0]
|
| 129 |
+
|
| 130 |
+
# Procesamiento con spaCy si está disponible
|
| 131 |
+
entities = []
|
| 132 |
+
if self.nlp:
|
| 133 |
+
doc = self.nlp(segment['text'])
|
| 134 |
+
entities = [(ent.text, ent.label_) for ent in doc.ents]
|
| 135 |
+
|
| 136 |
+
segments.append({
|
| 137 |
+
'start': segment['start'],
|
| 138 |
+
'end': segment['end'],
|
| 139 |
+
'text': segment['text'],
|
| 140 |
+
'confidence': segment.get('avg_logprob', 0),
|
| 141 |
+
'sentiment': sentiment,
|
| 142 |
+
'emotion': emotion,
|
| 143 |
+
'entities': entities,
|
| 144 |
+
'words': segment.get('words', [])
|
| 145 |
+
})
|
| 146 |
+
|
| 147 |
+
return {
|
| 148 |
+
'language': result['language'],
|
| 149 |
+
'segments': segments,
|
| 150 |
+
'full_text': result['text']
|
| 151 |
+
}
|
| 152 |
+
|
| 153 |
+
def intelligent_translation(self, transcription, target_language):
|
| 154 |
+
"""Traducción inteligente con múltiples modelos"""
|
| 155 |
+
print(f"🌍 Traduciendo a {target_language}...")
|
| 156 |
+
|
| 157 |
+
translated_segments = []
|
| 158 |
+
|
| 159 |
+
for segment in transcription['segments']:
|
| 160 |
+
original_text = segment['text']
|
| 161 |
+
|
| 162 |
+
# Traducción con Google Translate (más precisa)
|
| 163 |
+
try:
|
| 164 |
+
google_translation = self.google_translator.translate(
|
| 165 |
+
original_text,
|
| 166 |
+
dest=target_language
|
| 167 |
+
).text
|
| 168 |
+
except:
|
| 169 |
+
google_translation = original_text
|
| 170 |
+
|
| 171 |
+
# Preservar entidades nombradas
|
| 172 |
+
final_translation = google_translation
|
| 173 |
+
if segment['entities']:
|
| 174 |
+
for entity_text, entity_type in segment['entities']:
|
| 175 |
+
if entity_type in ['PERSON', 'ORG', 'GPE']:
|
| 176 |
+
final_translation = final_translation.replace(
|
| 177 |
+
entity_text.lower(), entity_text
|
| 178 |
+
)
|
| 179 |
+
|
| 180 |
+
translated_segments.append({
|
| 181 |
+
**segment,
|
| 182 |
+
'translated_text': final_translation,
|
| 183 |
+
'original_text': original_text
|
| 184 |
+
})
|
| 185 |
+
|
| 186 |
+
return translated_segments
|
| 187 |
+
|
| 188 |
+
def generate_smart_subtitles(self, segments, video_duration):
|
| 189 |
+
"""Genera subtítulos inteligentes con formato optimizado"""
|
| 190 |
+
print("📝 Generando subtítulos inteligentes...")
|
| 191 |
+
|
| 192 |
+
subtitles = []
|
| 193 |
+
|
| 194 |
+
for i, segment in enumerate(segments):
|
| 195 |
+
# Calcular duración óptima del subtítulo
|
| 196 |
+
duration = segment['end'] - segment['start']
|
| 197 |
+
text = segment.get('translated_text', segment['text'])
|
| 198 |
+
|
| 199 |
+
# Dividir texto largo en múltiples subtítulos
|
| 200 |
+
max_chars = 42 # Máximo caracteres por línea
|
| 201 |
+
max_lines = 2 # Máximo líneas por subtítulo
|
| 202 |
+
|
| 203 |
+
words = text.split()
|
| 204 |
+
lines = []
|
| 205 |
+
current_line = ""
|
| 206 |
+
|
| 207 |
+
for word in words:
|
| 208 |
+
if len(current_line + " " + word) <= max_chars:
|
| 209 |
+
current_line += (" " + word) if current_line else word
|
| 210 |
+
else:
|
| 211 |
+
if current_line:
|
| 212 |
+
lines.append(current_line)
|
| 213 |
+
current_line = word
|
| 214 |
+
|
| 215 |
+
if len(lines) >= max_lines:
|
| 216 |
+
break
|
| 217 |
+
|
| 218 |
+
if current_line:
|
| 219 |
+
lines.append(current_line)
|
| 220 |
+
|
| 221 |
+
# Crear subtítulo con formato
|
| 222 |
+
subtitle_text = "\n".join(lines[:max_lines])
|
| 223 |
+
|
| 224 |
+
# Aplicar estilo basado en emoción
|
| 225 |
+
emotion_label = segment['emotion']['label']
|
| 226 |
+
color = self.get_emotion_color(emotion_label)
|
| 227 |
+
|
| 228 |
+
subtitles.append({
|
| 229 |
+
'start': segment['start'],
|
| 230 |
+
'end': segment['end'],
|
| 231 |
+
'text': subtitle_text,
|
| 232 |
+
'emotion': emotion_label,
|
| 233 |
+
'color': color,
|
| 234 |
+
'confidence': segment['confidence']
|
| 235 |
+
})
|
| 236 |
+
|
| 237 |
+
return subtitles
|
| 238 |
+
|
| 239 |
+
def get_emotion_color(self, emotion):
|
| 240 |
+
"""Asigna colores basados en emociones"""
|
| 241 |
+
emotion_colors = {
|
| 242 |
+
'joy': 'yellow',
|
| 243 |
+
'sadness': 'blue',
|
| 244 |
+
'anger': 'red',
|
| 245 |
+
'fear': 'purple',
|
| 246 |
+
'surprise': 'orange',
|
| 247 |
+
'disgust': 'green',
|
| 248 |
+
'neutral': 'white'
|
| 249 |
+
}
|
| 250 |
+
return emotion_colors.get(emotion.lower(), 'white')
|
| 251 |
+
|
| 252 |
+
def create_subtitle_video(self, video_path, subtitles, output_path):
|
| 253 |
+
"""Crea video con subtítulos integrados"""
|
| 254 |
+
print("🎬 Creando video con subtítulos...")
|
| 255 |
+
|
| 256 |
+
video = VideoFileClip(video_path)
|
| 257 |
+
subtitle_clips = []
|
| 258 |
+
|
| 259 |
+
for subtitle in subtitles:
|
| 260 |
+
# Crear clip de texto con estilo
|
| 261 |
+
txt_clip = TextClip(
|
| 262 |
+
subtitle['text'],
|
| 263 |
+
fontsize=24,
|
| 264 |
+
font='Arial-Bold',
|
| 265 |
+
color=subtitle['color'],
|
| 266 |
+
stroke_color='black',
|
| 267 |
+
stroke_width=2
|
| 268 |
+
).set_position(('center', 'bottom')).set_duration(
|
| 269 |
+
subtitle['end'] - subtitle['start']
|
| 270 |
+
).set_start(subtitle['start'])
|
| 271 |
+
|
| 272 |
+
subtitle_clips.append(txt_clip)
|
| 273 |
+
|
| 274 |
+
# Componer video final
|
| 275 |
+
final_video = CompositeVideoClip([video] + subtitle_clips)
|
| 276 |
+
final_video.write_videofile(
|
| 277 |
+
output_path,
|
| 278 |
+
codec='libx264',
|
| 279 |
+
audio_codec='aac',
|
| 280 |
+
verbose=False,
|
| 281 |
+
logger=None
|
| 282 |
+
)
|
| 283 |
+
|
| 284 |
+
video.close()
|
| 285 |
+
final_video.close()
|
| 286 |
+
|
| 287 |
+
return output_path
|
| 288 |
+
|
| 289 |
+
def export_subtitle_formats(self, subtitles, base_path):
|
| 290 |
+
"""Exporta subtítulos en múltiples formatos"""
|
| 291 |
+
formats = {}
|
| 292 |
+
|
| 293 |
+
# Formato SRT
|
| 294 |
+
srt_path = f"{base_path}.srt"
|
| 295 |
+
with open(srt_path, 'w', encoding='utf-8') as f:
|
| 296 |
+
for i, sub in enumerate(subtitles, 1):
|
| 297 |
+
start_time = self.seconds_to_srt_time(sub['start'])
|
| 298 |
+
end_time = self.seconds_to_srt_time(sub['end'])
|
| 299 |
+
f.write(f"{i}\n{start_time} --> {end_time}\n{sub['text']}\n\n")
|
| 300 |
+
formats['srt'] = srt_path
|
| 301 |
+
|
| 302 |
+
# Formato VTT
|
| 303 |
+
vtt_path = f"{base_path}.vtt"
|
| 304 |
+
with open(vtt_path, 'w', encoding='utf-8') as f:
|
| 305 |
+
f.write("WEBVTT\n\n")
|
| 306 |
+
for sub in subtitles:
|
| 307 |
+
start_time = self.seconds_to_vtt_time(sub['start'])
|
| 308 |
+
end_time = self.seconds_to_vtt_time(sub['end'])
|
| 309 |
+
f.write(f"{start_time} --> {end_time}\n{sub['text']}\n\n")
|
| 310 |
+
formats['vtt'] = vtt_path
|
| 311 |
+
|
| 312 |
+
# Formato JSON con metadatos
|
| 313 |
+
json_path = f"{base_path}.json"
|
| 314 |
+
with open(json_path, 'w', encoding='utf-8') as f:
|
| 315 |
+
json.dump(subtitles, f, indent=2, ensure_ascii=False)
|
| 316 |
+
formats['json'] = json_path
|
| 317 |
+
|
| 318 |
+
return formats
|
| 319 |
+
|
| 320 |
+
def seconds_to_srt_time(self, seconds):
|
| 321 |
+
"""Convierte segundos a formato SRT"""
|
| 322 |
+
td = timedelta(seconds=seconds)
|
| 323 |
+
hours, remainder = divmod(td.total_seconds(), 3600)
|
| 324 |
+
minutes, seconds = divmod(remainder, 60)
|
| 325 |
+
milliseconds = int((seconds % 1) * 1000)
|
| 326 |
+
return f"{int(hours):02d}:{int(minutes):02d}:{int(seconds):02d},{milliseconds:03d}"
|
| 327 |
+
|
| 328 |
+
def seconds_to_vtt_time(self, seconds):
|
| 329 |
+
"""Convierte segundos a formato VTT"""
|
| 330 |
+
td = timedelta(seconds=seconds)
|
| 331 |
+
hours, remainder = divmod(td.total_seconds(), 3600)
|
| 332 |
+
minutes, seconds = divmod(remainder, 60)
|
| 333 |
+
milliseconds = int((seconds % 1) * 1000)
|
| 334 |
+
return f"{int(hours):02d}:{int(minutes):02d}:{int(seconds):02d}.{milliseconds:03d}"
|
| 335 |
+
|
| 336 |
+
def process_video(self, video_file, target_language="es", include_emotions=True):
|
| 337 |
+
"""Procesa video completo para generar subtítulos"""
|
| 338 |
+
if video_file is None:
|
| 339 |
+
return None, None, "Por favor sube un video"
|
| 340 |
+
|
| 341 |
+
try:
|
| 342 |
+
print("🎯 Iniciando procesamiento con ZenVision...")
|
| 343 |
+
|
| 344 |
+
# 1. Extraer características de audio
|
| 345 |
+
audio_features = self.extract_audio_features(video_file.name)
|
| 346 |
+
|
| 347 |
+
# 2. Transcripción avanzada
|
| 348 |
+
transcription = self.advanced_transcription(audio_features)
|
| 349 |
+
|
| 350 |
+
# 3. Traducción inteligente
|
| 351 |
+
if target_language != transcription['language']:
|
| 352 |
+
segments = self.intelligent_translation(transcription, target_language)
|
| 353 |
+
else:
|
| 354 |
+
segments = transcription['segments']
|
| 355 |
+
|
| 356 |
+
# 4. Generar subtítulos inteligentes
|
| 357 |
+
subtitles = self.generate_smart_subtitles(segments, audio_features['duration'])
|
| 358 |
+
|
| 359 |
+
# 5. Crear video con subtítulos
|
| 360 |
+
output_video_path = tempfile.mktemp(suffix=".mp4")
|
| 361 |
+
self.create_subtitle_video(video_file.name, subtitles, output_video_path)
|
| 362 |
+
|
| 363 |
+
# 6. Exportar formatos de subtítulos
|
| 364 |
+
subtitle_base_path = tempfile.mktemp()
|
| 365 |
+
subtitle_formats = self.export_subtitle_formats(subtitles, subtitle_base_path)
|
| 366 |
+
|
| 367 |
+
# Estadísticas del procesamiento
|
| 368 |
+
stats = {
|
| 369 |
+
'language_detected': transcription['language'],
|
| 370 |
+
'total_segments': len(subtitles),
|
| 371 |
+
'duration': audio_features['duration'],
|
| 372 |
+
'avg_confidence': np.mean([s['confidence'] for s in segments]),
|
| 373 |
+
'emotions_detected': len(set([s['emotion']['label'] for s in segments]))
|
| 374 |
+
}
|
| 375 |
+
|
| 376 |
+
status_msg = f"""✅ Procesamiento completado con ZenVision!
|
| 377 |
+
|
| 378 |
+
📊 Estadísticas:
|
| 379 |
+
• Idioma detectado: {stats['language_detected']}
|
| 380 |
+
• Segmentos generados: {stats['total_segments']}
|
| 381 |
+
• Duración: {stats['duration']:.1f}s
|
| 382 |
+
• Confianza promedio: {stats['avg_confidence']:.2f}
|
| 383 |
+
• Emociones detectadas: {stats['emotions_detected']}
|
| 384 |
+
|
| 385 |
+
🎯 Tecnologías utilizadas:
|
| 386 |
+
• Whisper Large-v2 (Transcripción)
|
| 387 |
+
• BERT Multilingual (Embeddings)
|
| 388 |
+
• RoBERTa (Análisis de sentimientos)
|
| 389 |
+
• DistilRoBERTa (Detección de emociones)
|
| 390 |
+
• Google Translate (Traducción)
|
| 391 |
+
• OpenCV + MoviePy (Procesamiento de video)
|
| 392 |
+
• Librosa (Análisis de audio)
|
| 393 |
+
• spaCy (NLP avanzado)
|
| 394 |
+
"""
|
| 395 |
+
|
| 396 |
+
return output_video_path, subtitle_formats['srt'], status_msg
|
| 397 |
+
|
| 398 |
+
except Exception as e:
|
| 399 |
+
return None, None, f"❌ Error en ZenVision: {str(e)}"
|
| 400 |
+
|
| 401 |
+
# Inicializar ZenVision
|
| 402 |
+
print("🚀 Inicializando ZenVision Model...")
|
| 403 |
+
zenvision = ZenVisionModel()
|
| 404 |
+
|
| 405 |
+
# Interfaz Gradio
|
| 406 |
+
with gr.Blocks(title="ZenVision - AI Subtitle Generator", theme=gr.themes.Soft()) as demo:
|
| 407 |
+
gr.HTML("""
|
| 408 |
+
<div style="text-align: center; padding: 20px;">
|
| 409 |
+
<h1>🎬 ZenVision AI Subtitle Generator</h1>
|
| 410 |
+
<p style="font-size: 18px; color: #666;">
|
| 411 |
+
Modelo avanzado de subtitulado automático con IA<br>
|
| 412 |
+
<strong>Desarrollado por el equipo ZenVision</strong>
|
| 413 |
+
</p>
|
| 414 |
+
<p style="font-size: 14px; color: #888;">
|
| 415 |
+
Modelo de 3GB+ • Whisper • BERT • RoBERTa • OpenCV • Librosa • spaCy
|
| 416 |
+
</p>
|
| 417 |
+
</div>
|
| 418 |
+
""")
|
| 419 |
+
|
| 420 |
+
with gr.Row():
|
| 421 |
+
with gr.Column(scale=1):
|
| 422 |
+
gr.Markdown("### 📤 Entrada")
|
| 423 |
+
video_input = gr.Video(label="Subir Video", height=300)
|
| 424 |
+
|
| 425 |
+
with gr.Row():
|
| 426 |
+
language_dropdown = gr.Dropdown(
|
| 427 |
+
choices=[
|
| 428 |
+
("Español", "es"),
|
| 429 |
+
("English", "en"),
|
| 430 |
+
("Français", "fr"),
|
| 431 |
+
("Deutsch", "de"),
|
| 432 |
+
("Italiano", "it"),
|
| 433 |
+
("Português", "pt"),
|
| 434 |
+
("中文", "zh"),
|
| 435 |
+
("日本語", "ja"),
|
| 436 |
+
("한국어", "ko"),
|
| 437 |
+
("Русский", "ru")
|
| 438 |
+
],
|
| 439 |
+
value="es",
|
| 440 |
+
label="Idioma de destino"
|
| 441 |
+
)
|
| 442 |
+
|
| 443 |
+
emotions_checkbox = gr.Checkbox(
|
| 444 |
+
label="Incluir análisis de emociones",
|
| 445 |
+
value=True
|
| 446 |
+
)
|
| 447 |
+
|
| 448 |
+
process_btn = gr.Button(
|
| 449 |
+
"🚀 Procesar con ZenVision",
|
| 450 |
+
variant="primary",
|
| 451 |
+
size="lg"
|
| 452 |
+
)
|
| 453 |
+
|
| 454 |
+
with gr.Column(scale=1):
|
| 455 |
+
gr.Markdown("### 📥 Resultados")
|
| 456 |
+
video_output = gr.Video(label="Video con Subtítulos", height=300)
|
| 457 |
+
subtitle_file = gr.File(label="Archivo de Subtítulos (.srt)")
|
| 458 |
+
|
| 459 |
+
with gr.Row():
|
| 460 |
+
status_output = gr.Textbox(
|
| 461 |
+
label="Estado del Procesamiento",
|
| 462 |
+
lines=15,
|
| 463 |
+
interactive=False
|
| 464 |
+
)
|
| 465 |
+
|
| 466 |
+
# Ejemplos
|
| 467 |
+
gr.Markdown("### 🎯 Características de ZenVision")
|
| 468 |
+
gr.HTML("""
|
| 469 |
+
<div style="display: grid; grid-template-columns: repeat(auto-fit, minmax(250px, 1fr)); gap: 15px; margin: 20px 0;">
|
| 470 |
+
<div style="padding: 15px; border: 1px solid #ddd; border-radius: 8px;">
|
| 471 |
+
<h4>🎤 Transcripción Avanzada</h4>
|
| 472 |
+
<p>Whisper Large-v2 con timestamps precisos y detección automática de idioma</p>
|
| 473 |
+
</div>
|
| 474 |
+
<div style="padding: 15px; border: 1px solid #ddd; border-radius: 8px;">
|
| 475 |
+
<h4>🌍 Traducción Inteligente</h4>
|
| 476 |
+
<p>Google Translate + preservación de entidades nombradas</p>
|
| 477 |
+
</div>
|
| 478 |
+
<div style="padding: 15px; border: 1px solid #ddd; border-radius: 8px;">
|
| 479 |
+
<h4>😊 Análisis Emocional</h4>
|
| 480 |
+
<p>Detección de emociones y sentimientos con colores adaptativos</p>
|
| 481 |
+
</div>
|
| 482 |
+
<div style="padding: 15px; border: 1px solid #ddd; border-radius: 8px;">
|
| 483 |
+
<h4>📝 Múltiples Formatos</h4>
|
| 484 |
+
<p>Exportación en SRT, VTT y JSON con metadatos completos</p>
|
| 485 |
+
</div>
|
| 486 |
+
</div>
|
| 487 |
+
""")
|
| 488 |
+
|
| 489 |
+
# Conectar funciones
|
| 490 |
+
process_btn.click(
|
| 491 |
+
fn=zenvision.process_video,
|
| 492 |
+
inputs=[video_input, language_dropdown, emotions_checkbox],
|
| 493 |
+
outputs=[video_output, subtitle_file, status_output]
|
| 494 |
+
)
|
| 495 |
+
|
| 496 |
+
if __name__ == "__main__":
|
| 497 |
+
demo.launch(
|
| 498 |
+
server_name="0.0.0.0",
|
| 499 |
+
server_port=7860,
|
| 500 |
+
share=True
|
| 501 |
+
)
|