Upload app.py

app.py

@@ -11,57 +11,13 @@ import whisper
 import pandas as pd
 from gtts import gTTS
 import re
+import io         # Necesario para el buffer en memoria
+import base64     # Necesario para la codificación
 
-# --- 0. CONFIGURACIÓN INICIAL ---
-try:
-    client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
-    api_key_found = True
-except TypeError:
-    api_key_found = False
+# --- (El resto de tu configuración y prompts se mantienen igual) ---
+# ...
 
-print("Loading Whisper for transcription...")
-whisper_model = whisper.load_model("base", device="cpu")
-print("Whisper model loaded.")
-
-
-# --- 1. DEFINICIÓN DE PROMPTS PARA LA IA ---
-# (Tus prompts completos van aquí...)
-CONVERSATION_SYSTEM_PROMPT = """..."""
-FINAL_EVALUATION_SYSTEM_PROMPT = """..."""
-SENTENCE_EVALUATION_SYSTEM_PROMPT = """
-You are an expert English language examiner...
-JSON Output Structure:
-{
-  "overall_score_100": integer,
-  "cefr_level": "string",
-  "holistic_feedback": { "strengths": "string", "areas_for_improvement": "string" },
-  "word_by_word_analysis": [ { "reference_word": "string", "spoken_word": "string", "word_score_100": integer, "correct_ipa": "string", "feedback_en": "string", "feedback_es": "string" } ]
-}
-"""
-
-# --- 2. FUNCIONES LÓGICAS ---
-
-def extract_word_level_features(audio_path):
-    try:
-        y, sr = librosa.load(audio_path, sr=16000)
-        result = whisper_model.transcribe(audio_path, word_timestamps=True, fp16=False)
-        if not result["segments"] or 'words' not in result["segments"][0]: return []
-        word_segments = result["segments"][0]["words"]
-        features_list = []
-        for segment in word_segments:
-            start_sample = int(segment['start'] * sr); end_sample = int(segment['end'] * sr)
-            word_audio = y[start_sample:end_sample]
-            rms_energy = np.mean(librosa.feature.rms(y=word_audio)) if len(word_audio) > 0 else 0
-            features_list.append({"word": segment['word'].strip(), "start": round(segment['start'], 2), "end": round(segment['end'], 2), "energy": round(float(rms_energy), 4)})
-        return features_list
-    except Exception as e:
-        print(f"Error during feature extraction: {e}"); return []
-
-def chat_interaction(audio_input, history_state):
-    # (Tu función de chat sin cambios va aquí...)
-    pass
-
-# --- CAMBIO: La función de evaluación de frase ahora devuelve MARKDOWN ---
+# --- CAMBIO: La función de evaluación ahora incrusta el audio en Base64 ---
 def run_sentence_evaluation(audio_input, reference_transcript):
     if not api_key_found: raise gr.Error("OpenAI API key not found.")
     if audio_input is None or not reference_transcript:
@@ -81,21 +37,26 @@ def run_sentence_evaluation(audio_input, reference_transcript):
         holistic_feedback_md = f"### Strengths\n{result['holistic_feedback']['strengths']}\n\n### Areas for Improvement\n{result['holistic_feedback']['areas_for_improvement']}"
         word_analysis_list = result['word_by_word_analysis']
         
-        # --- NUEVA LÓGICA: Construir una tabla en Markdown ---
+        # --- NUEVA LÓGICA: Construir tabla en Markdown con audio Base64 ---
         md_table = "| Reference Word | Spoken Word | Score | Feedback (EN) | Feedback (ES) | Reference Audio |\n"
         md_table += "| :--- | :--- | :---: | :--- | :--- | :---: |\n"
         
-        os.makedirs("reference_audio", exist_ok=True)
-        
         for index, item in enumerate(word_analysis_list):
             word_to_speak = item['reference_word']
-            safe_filename = re.sub(r'\W+', '', word_to_speak.lower())
-            audio_path = f"reference_audio/{index}_{safe_filename}.mp3"
             
             try:
-                tts = gTTS(text=word_to_speak, lang='en'); tts.save(audio_path)
-                # Embeber el audio usando una etiqueta HTML <audio>
-                audio_player = f'<audio src="file/{audio_path}" controls></audio>'
+                # 1. Generar audio en un buffer en memoria (no en un archivo)
+                tts = gTTS(text=word_to_speak, lang='en')
+                mp3_fp = io.BytesIO()
+                tts.write_to_fp(mp3_fp)
+                mp3_fp.seek(0)
+                
+                # 2. Codificar el audio en Base64
+                audio_base64 = base64.b64encode(mp3_fp.read()).decode('utf-8')
+                
+                # 3. Crear una etiqueta de audio con los datos incrustados (Data URI)
+                audio_player = f'<audio src="data:audio/mpeg;base64,{audio_base64}" controls></audio>'
+
             except Exception as e:
                 print(f"Error al generar TTS para '{word_to_speak}': {e}"); audio_player = "Error"
 
@@ -107,40 +68,20 @@ def run_sentence_evaluation(audio_input, reference_transcript):
         return 0, "Error", error_msg, ""
 
 
-# --- 3. INTERFAZ DE GRADIO CON PESTAÑAS (Con salida Markdown) ---
+# --- 3. INTERFAZ DE GRADIO CON PESTAÑAS (La definición no cambia, pero ahora recibe Markdown) ---
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("# 🇬🇧 AI English Speaking Practice & Assessment")
-    with gr.Tabs():
-        # PESTAÑA 1: CHAT AI (sin cambios)
-        with gr.TabItem("Práctica Conversacional (Chat AI)"):
-            # ... (Aquí va toda la definición de la interfaz de tu chatbot, sin cambios)
-            pass
-
-        # PESTAÑA 2: EVALUACIÓN POR FRASE
+    # ... (Todo el resto de tu interfaz, incluyendo la Pestaña 1 y 2, se mantiene exactamente igual)
+    # ...
         with gr.TabItem("Evaluación por Frase"):
-            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?"]
-            gr.Markdown("Choose a tongue twister or write your own sentence...")
-            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")
-            
-            # --- AJUSTE CLAVE: La salida ahora es un único componente Markdown ---
-            word_analysis_out_sentence = gr.Markdown(label="Phonetic Breakdown")
-
-            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])
+            # ... (Toda la definición de la interfaz de esta pestaña se mantiene igual)
+            # ...
+            word_analysis_out_sentence = gr.Markdown(label="Phonetic Breakdown") # <-- Esta salida ya está lista para recibir la tabla Markdown
+            # ...
+            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: print("\nFATAL: OpenAI API key not found.")