Update app.py

app.py

@@ -3,19 +3,22 @@ import base64
 import io
 from huggingface_hub import InferenceClient
 from gtts import gTTS
-import speech_recognition as sr
 from audiorecorder import audiorecorder
+import speech_recognition as sr
+from pydub import AudioSegment
 
 if "history" not in st.session_state:
     st.session_state.history = []
 
 def recognize_speech(audio_data, show_messages=True):
     recognizer = sr.Recognizer()
-    audio = sr.AudioData(audio_data, sample_rate=16000)
+    audio_recording = sr.AudioFile(audio_data)
+
+    with audio_recording as source:
+        audio = recognizer.record(source)
 
     try:
-        # Eliminar la clave de API de Google Cloud Speech
-        audio_text = recognizer.recognize_google(audio, language="es-us")
+        audio_text = recognizer.recognize_google(audio, language="es-ES")
         if show_messages:
             st.subheader("Texto Reconocido:")
             st.write(audio_text)
@@ -54,8 +57,6 @@ def generate(audio_text, history, temperature=None, max_new_tokens=512, top_p=0.
         repetition_penalty=repetition_penalty,
         do_sample=True,
         seed=42,
-        # Habilitar el uso de la caché del modelo
-        use_cache=True
     )
 
     formatted_prompt = format_prompt(audio_text, history)
@@ -66,37 +67,47 @@ def generate(audio_text, history, temperature=None, max_new_tokens=512, top_p=0.
         response += response_token.token.text
     
     response = ' '.join(response.split()).replace('</s>', '')
-    audio_file = text_to_speech(response)
+    audio_file = text_to_speech(response, speed=1.3)
     return response, audio_file
 
-def text_to_speech(text):
-    tts = gTTS(text, lang="es", tld="io", name="Alvaro")
-    audio_file = io.BytesIO()
-    tts.write_to_fp(audio_file)
-    audio_file.seek(0)
-    return audio_file
-
-def background_process(audio_data):
-    audio_text = recognize_speech(audio_data, show_messages=False)
-    if audio_text:
-        output, audio_file = generate(audio_text, history=st.session_state.history)
-        st.session_state.history.append((audio_text, output))
-        st.markdown(f"## Respuesta generada:\n{output}")
-        st.audio(audio_file, format="audio/mp3")
+def text_to_speech(text, speed=1.3):
+    tts = gTTS(text=text, lang='es')
+    audio_fp = io.BytesIO()
+    tts.write_to_fp(audio_fp)
+    audio_fp.seek(0)
+    audio = AudioSegment.from_file(audio_fp, format="mp3")
+    modified_speed_audio = audio.speedup(playback_speed=speed)
+    modified_audio_fp = io.BytesIO()
+    modified_speed_audio.export(modified_audio_fp, format="mp3")
+    modified_audio_fp.seek(0)
+    return modified_audio_fp
 
 def main():
-    audio_recorder = audiorecorder(dur=10, filename="audio.wav", listen_background=background_process, noise_reduction=True)
-    st.markdown("## Grabar audio")
-    st.markdown("Presiona el botón de abajo para grabar tu voz durante 10 segundos.")
-    record_state = audio_recorder.record()
-    if record_state:
-        st.info("Grabando...")
-    else:
-        st.success("Grabación completada.")
-    st.markdown("## Reproducir audio")
-    st.markdown("Presiona el botón de abajo para reproducir el audio grabado.")
-    play_state = audio_recorder.play()
-    if play_state:
-        st.info("Reproduciendo...")
-    else:
-        st.success("Reproducción finalizada.")
+    audio_data = audiorecorder("Habla para grabar", "Deteniendo la grabación...")
+
+    if not audio_data.empty():
+        st.audio(audio_data.export().read(), format="audio/wav")
+        audio_data.export("audio.wav", format="wav")
+        audio_text = recognize_speech("audio.wav")
+
+        if not st.session_state.history:
+            pre_prompt = "Te Llamarás Chaman 4.0 y tus respuestas serán sumamente breves."
+            output, _ = generate(pre_prompt, history=st.session_state.history)
+            st.session_state.history.append((pre_prompt, output))
+        
+        if audio_text:
+            output, audio_file = generate(audio_text, history=st.session_state.history)  
+
+            if audio_text:  
+                st.session_state.history.append((audio_text, output))  
+
+            if audio_file is not None:
+                st.markdown(
+                    f"""
+                    <audio autoplay="autoplay" controls="controls" src="data:audio/mp3;base64,{base64.b64encode(audio_file.read()).decode()}" type="audio/mp3" id="audio_player"></audio>
+                    """,
+                    unsafe_allow_html=True
+                )
+
+if __name__ == "__main__":
+    main()