Update app.py

app.py

@@ -5,105 +5,136 @@ import json
 import gspread
 from oauth2client.service_account import ServiceAccountCredentials
 from datetime import datetime
+from gtts import gTTS
+import tempfile
 
 # --- CONFIGURATION ---
 MODEL_K2H_REPO = "ankitklakra/kurukh-to-hindi"
 MODEL_H2K_REPO = "ankitklakra/hindi-to-kurukh"
-SHEET_NAME = "Kurukh_Feedback_Log" 
+SHEET_NAME = "Kurukh_Feedback_Log"
 
-# --- GOOGLE SHEETS SETUP ---
+# --- LOAD TEXT MODELS ---
+print("Loading Translation Models...")
+tokenizer = AutoTokenizer.from_pretrained("google/mt5-small")
+model_k2h = AutoModelForSeq2SeqLM.from_pretrained(MODEL_K2H_REPO)
+model_h2k = AutoModelForSeq2SeqLM.from_pretrained(MODEL_H2K_REPO)
+
+pipe_k2h = pipeline("text2text-generation", model=model_k2h, tokenizer=tokenizer)
+pipe_h2k = pipeline("text2text-generation", model=model_h2k, tokenizer=tokenizer)
+
+# --- LOAD AUDIO MODEL (WHISPER) ---
+print("Loading Voice Model...")
+
+asr_pipeline = pipeline("automatic-speech-recognition", model="openai/whisper-tiny")
+
+# --- GOOGLE SHEETS FUNCTION  ---
 def save_to_sheet(original, translation, correction, direction):
     try:
-        # 1. Get credentials from Hugging Face Secrets
         json_creds = os.getenv("GOOGLE_CREDENTIALS")
-        if not json_creds:
-            return "⚠️ Error: Google Credentials not found in Settings."
-        
+        if not json_creds: return "⚠️ Error: Credentials missing."
         creds_dict = json.loads(json_creds)
-        
-        # 2. Authenticate
         scope = ["https://spreadsheets.google.com/feeds", "https://www.googleapis.com/auth/drive"]
         creds = ServiceAccountCredentials.from_json_keyfile_dict(creds_dict, scope)
         client = gspread.authorize(creds)
-        
-        # 3. Open Sheet
         sheet = client.open(SHEET_NAME).sheet1
-        
-        # 4. Add Headers if empty
         if not sheet.get_all_values():
             sheet.append_row(["Timestamp", "Direction", "Original Text", "AI Translation", "User Correction"])
-            
-        # 5. Append Row
         sheet.append_row([str(datetime.now()), direction, original, translation, correction])
-        return "✅ Success! Saved to Google Sheets."
-        
+        return "✅ Saved to Google Sheets."
     except Exception as e:
-        return f"❌ Error saving to Sheet: {str(e)}"
-
-# --- LOAD RESOURCES ---
-print("Loading Tokenizer...")
-tokenizer = AutoTokenizer.from_pretrained("google/mt5-small")
+        return f"❌ Error: {str(e)}"
 
-print("Loading Models...")
-model_k2h = AutoModelForSeq2SeqLM.from_pretrained(MODEL_K2H_REPO)
-model_h2k = AutoModelForSeq2SeqLM.from_pretrained(MODEL_H2K_REPO)
+# --- CORE FUNCTIONS ---
 
-pipe_k2h = pipeline("text2text-generation", model=model_k2h, tokenizer=tokenizer)
-pipe_h2k = pipeline("text2text-generation", model=model_h2k, tokenizer=tokenizer)
+def speech_to_text(audio_path):
+    """Converts Microphone Audio -> Hindi/English Text"""
+    if audio_path is None:
+        return ""
+    print("Transcribing audio...")
+    text = asr_pipeline(audio_path)["text"]
+    return text
 
-# --- TRANSLATION FUNCTION ---
-def translate_text(text, direction):
+def text_to_speech(text, language="hi"):
+    """Converts Text -> Audio File"""
     if not text:
-        return ""
+        return None
+    try:
+        # Save audio to a temporary file
+        tts = gTTS(text=text, lang=language)
+        temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3")
+        tts.save(temp_file.name)
+        return temp_file.name
+    except:
+        return None
+
+def process_translation(text, audio_input, direction):
+    # 1. Handle Input Source (Audio vs Text)
+    if audio_input is not None:
+        # If user spoke, transcribe it first
+        original_text = speech_to_text(audio_input)
+    else:
+        original_text = text
+
+    if not original_text:
+        return "", "", None
+
+    # 2. Translate
     target_pipeline = pipe_k2h if direction == "Kurukh -> Hindi" else pipe_h2k
     try:
-        results = target_pipeline(text, max_length=128)
-        return results[0]['generated_text']
+        results = target_pipeline(original_text, max_length=128)
+        translated_text = results[0]['generated_text']
     except Exception as e:
-        return f"Error: {str(e)}"
+        return str(e), "", None
+
+    # 3. Handle Audio Output (Read aloud if Hindi)
+    audio_output = None
+    
+    if direction == "Kurukh -> Hindi":
+        audio_output = text_to_speech(translated_text, "hi")
+    
+    # Return: (Input Text shown to user), (Translation), (Audio File)
+    return original_text, translated_text, audio_output
 
 # --- THE UI ---
 with gr.Blocks() as demo:
     gr.Markdown("# 🇮🇳 AI Kurukh (Kurux) Translator")
-    gr.Markdown("### Preserving Tribal Languages with Artificial Intelligence")
+    gr.Markdown("### Voice-Enabled Tribal Language Preserver")
     
     with gr.Tabs():
-        # TAB 1: TRANSLATOR
-        with gr.TabItem("🗣️ Translator"):
-            with gr.Row():
-                direction = gr.Radio(["Kurukh -> Hindi", "Hindi -> Kurukh"], label="Translation Mode", value="Kurukh -> Hindi")
+        with gr.TabItem("🗣️ Voice & Text Translator"):
+            
             with gr.Row():
-                input_text = gr.Textbox(label="Input Text", placeholder="Type here...", lines=4)
-                output_text = gr.Textbox(label="AI Translation", lines=4, interactive=False)
+                direction = gr.Radio(["Kurukh -> Hindi", "Hindi -> Kurukh"], label="Mode", value="Kurukh -> Hindi")
             
-            translate_btn = gr.Button("Translate 🚀", variant="primary")
-            translate_btn.click(fn=translate_text, inputs=[input_text, direction], outputs=output_text)
+            # INPUT SECTION
+            with gr.Row():
+                with gr.Column():
+                    input_text = gr.Textbox(label="Type Here", placeholder="Or use the microphone below...", lines=3)
+                    input_audio = gr.Audio(sources=["microphone"], type="filepath", label="Or Speak (Hindi Only)")
+                    translate_btn = gr.Button("Translate 🚀", variant="primary")
+                
+                # OUTPUT SECTION
+                with gr.Column():
+                    output_text = gr.Textbox(label="Translation", lines=3, interactive=False)
+                    output_audio = gr.Audio(label="Listen to Translation (Hindi)", interactive=False)
             
-            gr.Examples(
-                examples=[
-                    ["निघै नामे इन्द्रा हिकै?", "Kurukh -> Hindi"],
-                    ["तुम्हारा नाम क्या है?", "Hindi -> Kurukh"]
-                ],
-                inputs=[input_text, direction]
+            # LOGIC
+            translate_btn.click(
+                fn=process_translation, 
+                inputs=[input_text, input_audio, direction], 
+                outputs=[input_text, output_text, output_audio] # Updates input box with transcribed text too!
             )
 
-        # TAB 2: GOOGLE SHEETS FEEDBACK
         with gr.TabItem("📝 Improve the AI"):
             gr.Markdown("### Help us improve!")
             with gr.Row():
                 fb_direction = gr.Radio(["Kurukh -> Hindi", "Hindi -> Kurukh"], label="Direction", value="Kurukh -> Hindi")
             with gr.Row():
-                fb_original = gr.Textbox(label="Original Text")
-                fb_ai_output = gr.Textbox(label="What the AI said (Optional)")
-                fb_user_correct = gr.Textbox(label="Your Correction (Required)", lines=2)
-            
-            submit_btn = gr.Button("Submit")
+                fb_original = gr.Textbox(label="Original")
+                fb_ai_output = gr.Textbox(label="AI Said")
+                fb_user_correct = gr.Textbox(label="Correction", lines=2)
+            submit_btn = gr.Button("Submit to Google Sheets ✅")
             status_lbl = gr.Label(label="Status")
-            
-            submit_btn.click(
-                fn=save_to_sheet,
-                inputs=[fb_original, fb_ai_output, fb_user_correct, fb_direction],
-                outputs=status_lbl
-            )
+            submit_btn.click(fn=save_to_sheet, inputs=[fb_original, fb_ai_output, fb_user_correct, fb_direction], outputs=status_lbl)
 
 demo.launch()