Upload app.py

app.py

@@ -18,175 +18,136 @@ except TypeError:
     api_key_found = False
 
 print("Loading Whisper for transcription...")
-# Usamos el modelo 'base' que es un buen compromiso entre velocidad y precisión
 whisper_model = whisper.load_model("base", device="cpu") 
 print("Whisper model loaded.")
 
 
 # --- 1. DEFINICIÓN DE PROMPTS PARA LA IA ---
 
-# Prompt para mantener la conversación
-CONVERSATION_SYSTEM_PROMPT = """
-You are a friendly and encouraging English language tutor named Alex. 
-A student will speak to you. Your task is to keep a natural, simple conversation going.
-1. Briefly analyze the user's previous response to estimate their CEFR level (A1, A2, B1, etc.).
-2. Formulate a simple, open-ended follow-up question that is appropriate for THAT estimated level.
-3. Your entire response must be a single, short paragraph in natural, conversational English. DO NOT use JSON.
-"""
-
-# Prompt para la evaluación final
-FINAL_EVALUATION_SYSTEM_PROMPT = """
-You are an expert English language examiner providing a final report. Analyze the entire conversation history provided.
+# Prompt para la conversación
+CONVERSATION_SYSTEM_PROMPT = "..." # (Mantener el prompt de conversación que ya tienes)
 
-Your task is to return a single, valid JSON object with the following structure. Do not include any text outside this JSON object.
+# Prompt para la evaluación final de la conversación
+FINAL_EVALUATION_SYSTEM_PROMPT = "..." # (Mantener el prompt de evaluación final que ya tienes)
 
+# Prompt para la evaluación de una sola frase
+SENTENCE_EVALUATION_SYSTEM_PROMPT = """
+You are an expert English language examiner specializing in phonetics. Your task is to provide a detailed, diagnostic assessment of a student's spoken English based on a reference sentence and detailed word-level audio analysis. Your entire response MUST be in English. You must return a single, valid JSON object with the following structure. Do not include any text outside of this JSON object.
 JSON Output Structure:
 {
-  "cefr_level": "string (e.g., A2, B1)",
-  "feedback_en": {
-    "strengths": "string (A paragraph summarizing the student's strong points in pronunciation, vocabulary, and fluency.)",
-    "areas_for_improvement": "string (A paragraph detailing the main patterns of error and what to focus on.)",
-    "word_by_word_feedback": [
-      {"word": "string", "feedback": "string (Specific phonetic or usage feedback.)"}
-    ]
+  "overall_score_100": integer,
+  "cefr_level": "string (A1, A2, B1, B2, C1, or C2)",
+  "holistic_feedback": {
+    "strengths": "string (A paragraph in English summarizing strong points.)",
+    "areas_for_improvement": "string (A paragraph in English detailing main error patterns.)"
   },
-  "feedback_es": {
-    "fortalezas": "string (Un párrafo resumiendo los puntos fuertes del estudiante en pronunciación, vocabulario y fluidez.)",
-    "areas_a_mejorar": "string (Un párrafo detallando los patrones de error principales y en qué enfocarse.)",
-    "feedback_por_palabra": [
-      {"palabra": "string", "feedback": "string (Retroalimentación fonética o de uso específica.)"}
-    ]
-  }
+  "word_by_word_analysis": [
+    {
+      "reference_word": "string",
+      "spoken_word": "string",
+      "word_score_100": integer,
+      "correct_ipa": "string",
+      "feedback": "string"
+    }
+  ]
 }
 """
 
 
 # --- 2. FUNCIONES LÓGICAS ---
 
-def transcribe_audio(audio_input):
-    """Transcribe el audio usando la API de Whisper de OpenAI."""
-    if audio_input is None:
-        return ""
+# Función para la Pestaña "Práctica Conversacional"
+def chat_interaction(audio_input, history_state):
+    # ... (Mantener la función 'chat_interaction' completa que ya tienes)
+    pass # Placeholder for your existing chat_interaction function
+
+# Función para la Pestaña "Evaluación por Frase"
+def run_sentence_evaluation(audio_input, reference_transcript):
+    # This is the 'run_evaluation' function from your previous, single-tab app
+    if not api_key_found: raise gr.Error("OpenAI API key not found.")
+    if audio_input is None or not reference_transcript:
+        return 0, "N/A", "Please provide both an audio file and the reference text.", None
+
     sr, y = audio_input
     temp_audio_path = "temp_audio.wav"
     sf.write(temp_audio_path, y, sr)
-    with open(temp_audio_path, "rb") as audio_file:
-        transcript = client.audio.transcriptions.create(
-            model="whisper-1", 
-            file=audio_file
-        ).text
-    return transcript
-
-def chat_interaction(audio_input, history_state):
-    """
-    Gestiona una vuelta de la conversación.
-    """
-    if not api_key_found: raise gr.Error("OpenAI API key not found.")
-    if audio_input is None: return history_state, history_state, "", ""
 
-    # 1. Transcribir el audio del usuario
-    user_text = transcribe_audio(audio_input)
+    # (La lógica de 'extract_word_level_features' y la llamada a la API va aquí)
+    # For brevity, this part is condensed. Use the full function from your previous script.
     
-    # 2. Actualizar el historial con el mensaje del usuario
-    history_state.append({"role": "user", "content": user_text})
+    # Placeholder for the actual API call logic
+    word_analysis_df = pd.DataFrame({
+        "Reference Word": reference_transcript.split(),
+        "Spoken Word": reference_transcript.split(),
+        "Score": [np.random.randint(80, 100) for _ in reference_transcript.split()],
+        "Correct IPA": ["..."], "Feedback": ["..."]
+    })
+    holistic_feedback_md = "### Strengths\nExcellent clarity.\n\n### Areas for Improvement\nWork on sentence intonation."
     
-    # Formatear para el chatbot de Gradio
-    chat_display = []
-    for i, msg in enumerate(history_state):
-        if msg['role'] == 'user':
-            chat_display.append((msg['content'], None))
-        elif msg['role'] == 'assistant':
-            if chat_display and chat_display[-1][1] is None:
-                chat_display[-1] = (chat_display[-1][0], msg['content'])
-
-    # 3. Decidir si continuar la conversación o dar el reporte final
-    if len(history_state) < 9: # 1 system + 4 pares de user/assistant
-        # --- Continuar conversación ---
-        messages_to_send = [{"role": "system", "content": CONVERSATION_SYSTEM_PROMPT}] + history_state
-        
-        response = client.chat.completions.create(
-            model="gpt-4o",
-            messages=messages_to_send,
-            temperature=0.7
-        )
-        ai_response = response.choices[0].message.content
-        history_state.append({"role": "assistant", "content": ai_response})
-        chat_display[-1] = (chat_display[-1][0], ai_response)
-        
-        return chat_display, history_state, gr.Markdown(visible=False), gr.Markdown(visible=False)
+    return 92, "B2", holistic_feedback_md, gr.DataFrame(value=word_analysis_df)
 
-    else:
-        # --- Generar evaluación final ---
-        print("Generating final evaluation...")
-        messages_to_send = [{"role": "system", "content": FINAL_EVALUATION_SYSTEM_PROMPT}] + history_state
-        
-        response = client.chat.completions.create(
-            model="gpt-4o",
-            response_format={"type": "json_object"},
-            messages=messages_to_send
-        )
-        
-        try:
-            result = json.loads(response.choices[0].message.content)
-            
-            # Formatear el feedback en Inglés
-            fb_en = result.get('feedback_en', {})
-            md_en = f"## Final Report (CEFR Level: {result.get('cefr_level', 'N/A')})\n"
-            md_en += f"### Strengths\n{fb_en.get('strengths', '')}\n"
-            md_en += f"### Areas for Improvement\n{fb_en.get('areas_for_improvement', '')}\n"
-            md_en += "### Word-by-Word Feedback\n"
-            for item in fb_en.get('word_by_word_feedback', []):
-                md_en += f"- **{item['word']}**: {item['feedback']}\n"
-            
-            # Formatear el feedback en Español
-            fb_es = result.get('feedback_es', {})
-            md_es = f"## Reporte Final (Nivel MCERL: {result.get('cefr_level', 'N/A')})\n"
-            md_es += f"### Fortalezas\n{fb_es.get('fortalezas', '')}\n"
-            md_es += f"### Áreas a Mejorar\n{fb_es.get('areas_a_mejorar', '')}\n"
-            md_es += "### Retroalimentación por Palabra\n"
-            for item in fb_es.get('feedback_por_palabra', []):
-                md_es += f"- **{item['palabra']}**: {item['feedback']}\n"
-            
-            # Mensaje final para el chat
-            final_message = "Thank you for the conversation! Here is your final report."
-            chat_display[-1] = (chat_display[-1][0], final_message)
-            
-            return chat_display, history_state, gr.Markdown(value=md_en, visible=True), gr.Markdown(value=md_es, visible=True)
 
-        except (json.JSONDecodeError, KeyError) as e:
-            print(f"Error parsing final report: {e}")
-            return chat_display, history_state, gr.Markdown(value="Error generating report.", visible=True), gr.Markdown(visible=False)
-
-# --- 3. INTERFAZ DE GRADIO ---
+# --- 3. INTERFAZ DE GRADIO CON PESTAÑAS ---
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown("# 🇬🇧 AI English Speaking Practice & Assessment")
     
     with gr.Tabs():
+        # --- Pestaña 1: Chat AI ---
         with gr.TabItem("Práctica Conversacional (Chat AI)"):
             with gr.Row():
                 with gr.Column(scale=2):
-                    chatbot = gr.Chatbot(
-                        value=[(None, "Hi there! I'm Alex. How are you doing today?")],
-                        label="Conversation with your AI Tutor"
-                    )
+                    chatbot = gr.Chatbot(value=[(None, "Hi there! I'm Alex. How are you doing today?")], label="Conversation with your AI Tutor")
                     audio_in_chat = gr.Audio(sources=["microphone"], type="numpy", label="Record your response")
                 with gr.Column(scale=1):
                     gr.Markdown("### Final Report")
                     feedback_en_out = gr.Markdown(label="English Feedback", visible=False)
                     feedback_es_out = gr.Markdown(label="Retroalimentación en Español", visible=False)
-
-            # Estado para guardar el historial de la conversación
             history = gr.State([]) 
-            
             audio_in_chat.stop_recording(
                 fn=chat_interaction,
                 inputs=[audio_in_chat, history],
                 outputs=[chatbot, history, feedback_en_out, feedback_es_out]
             )
 
+        # --- Pestaña 2: Evaluación por Frase ---
         with gr.TabItem("Evaluación por Frase"):
-            gr.Markdown("This is a placeholder for the original sentence evaluation tool.")
-            # Aquí pegarías la interfaz de la herramienta anterior si quisieras combinar ambas.
+            TONGUE_TWISTERS = [
+                "Peter Piper picked a peck of pickled peppers.",
+                "She sells seashells by the seashore.",
+                "How much wood would a woodchuck chuck if a woodchuck could chuck wood?",
+                "Betty Botter bought some butter but she said the butter’s bitter.",
+                "A proper copper coffee pot."
+            ]
+            
+            gr.Markdown("Choose a tongue twister or write your own sentence. Record yourself, and our AI examiner will provide a detailed diagnostic report.")
+            
+            tongue_twister_selector = gr.Dropdown(choices=TONGUE_TWISTERS, label="Or Choose a Tongue Twister to Practice")
+            
+            with gr.Row():
+                with gr.Column(scale=1):
+                    audio_in_sentence = gr.Audio(sources=["microphone"], type="numpy", label="1. Record Your Voice")
+                    text_in_sentence = gr.Textbox(lines=3, label="2. Reference Sentence", value=TONGUE_TWISTERS[0])
+                    submit_btn_sentence = gr.Button("Get Assessment", variant="primary")
+                
+                with gr.Column(scale=2):
+                    gr.Markdown("### Assessment Summary")
+                    with gr.Row():
+                        score_out_sentence = gr.Number(label="Overall Score (0-100)", interactive=False)
+                        level_out_sentence = gr.Textbox(label="Estimated CEFR Level", interactive=False)
+                    holistic_feedback_out_sentence = gr.Markdown(label="Examiner's Feedback")
+            
+            gr.Markdown("--- \n ### Detailed Word-by-Word Analysis")
+            word_analysis_out_sentence = gr.DataFrame(headers=["Reference Word", "Spoken Word", "Score", "Correct IPA", "Feedback"], label="Phonetic Breakdown", wrap=True)
+
+            def update_text(choice):
+                return gr.Textbox(value=choice)
+            tongue_twister_selector.change(fn=update_text, inputs=tongue_twister_selector, outputs=text_in_sentence)
+
+            submit_btn_sentence.click(
+                fn=run_sentence_evaluation,
+                inputs=[audio_in_sentence, text_in_sentence],
+                outputs=[score_out_sentence, level_out_sentence, holistic_feedback_out_sentence, word_analysis_out_sentence]
+            )
 
 if __name__ == "__main__":
     if not api_key_found: