lang detection, translator, offensive detection

app.py

@@ -1,31 +1,67 @@
 import gradio as gr
 from transformers import pipeline
 
-# Load language detection model from Hugging Face
+# Load language detection model
 lang_classifier = pipeline("text-classification", model="papluca/xlm-roberta-base-language-detection")
 
-# Load hate speech detection model from Hugging Face
-hate_classifier = pipeline("text-classification", model="Hate-speech-CNERG/dehatebert-mono-english")
+# Load translation model (multi-language to English)
+translator = pipeline("translation", model="facebook/nllb-200-distilled-600M")
+
+# Load hate speech detection model
+offensive_classifier = pipeline("text-classification", model="cardiffnlp/twitter-roberta-base-offensive")
+
+# Mapping from ISO 639-1 to NLLB-200 language codes
+LANGUAGE_CODES = {
+    "en": "eng_Latn", "fr": "fra_Latn", "es": "spa_Latn", "de": "deu_Latn",
+    "bg": "bul_Cyrl", "ru": "rus_Cyrl", "it": "ita_Latn", "zh": "zho_Hans", 
+    "ar": "arb_Arab", "pt": "por_Latn", "nl": "nld_Latn", "hi": "hin_Deva"
+}
 
 def analyze_text(text):
     if not text.strip():
         return {"error": "No text provided"}, {"error": "No text provided"}
 
+    # Detect language
     lang_result = lang_classifier(text)
-    hate_result = hate_classifier(text)
+    detected_language = lang_result[0]['label']
+    language_confidence = lang_result[0]['score']
+    
+    # Convert detected language to NLLB-200 format
+    detected_language_nllb = LANGUAGE_CODES.get(detected_language, "eng_Latn")
+    
+    # Translate if not English
+    translated_text = text
+    if detected_language_nllb != "eng_Latn":
+        translation_result = translator(text, src_lang=detected_language_nllb, tgt_lang="eng_Latn")
+        translated_text = translation_result[0]['translation_text']
+    
+    # Detect hate speech using the translated text
+    hate_result = offensive_classifier(translated_text)
 
-    language_output = {"language": lang_result[0]['label'], "confidence": lang_result[0]['score']}
-    hate_output = {"label": hate_result[0]['label'], "score": hate_result[0]['score']}
+    language_output = {
+        "language": detected_language,
+        "confidence": language_confidence,
+        "original_text": text,
+        "translated_text": translated_text if detected_language_nllb != "eng_Latn" else "Already in English"
+    }
+    
+    hate_output = {
+        "label": hate_result[0]['label'],
+        "score": hate_result[0]['score']
+    }
 
     return language_output, hate_output
 
 # Define the Gradio interface
 iface = gr.Interface(
-    fn=analyze_text,  # Use the combined function
+    fn=analyze_text,
     inputs=gr.Textbox(label="Enter text"),
-    outputs=[gr.JSON(label="Language Detection"), gr.JSON(label="Hate Speech Detection")],
-    title="Language & Hate Speech Detection",
-    description="Enter text to detect its language and check for hate speech."
+    outputs=[
+        gr.JSON(label="Language Detection & Translation"),
+        gr.JSON(label="Hate Speech Detection")
+    ],
+    title="Language & Hate Speech Detection with Translation",
+    description="Enter text to detect its language, translate it to English if needed, and check for hate speech."
 )
 
 # Launch the Gradio app