Update app.py

app.py

@@ -1,11 +1,9 @@
 import streamlit as st
-import base64
 import io
 from huggingface_hub import InferenceClient
 from gtts import gTTS
 from audiorecorder import audiorecorder
 import speech_recognition as sr
-from pydub import AudioSegment
 
 pre_prompt_text = "eres una IA conductual, tus respuestas serán breves."
 
@@ -17,10 +15,16 @@ if "pre_prompt_sent" not in st.session_state:
 
 def recognize_speech(audio_data, show_messages=True):
     recognizer = sr.Recognizer()
-    audio_recording = sr.AudioFile(audio_data)
+
+    audio_filename = "temp_audio_file.wav"
+    with open(audio_filename, "wb") as f:
+        f.write(audio_data.read())
+
+    audio_recording = sr.AudioFile(audio_filename)
 
     with audio_recording as source:
         audio = recognizer.record(source)
+
     try:
         audio_text = recognizer.recognize_google(audio, language="es-ES")
         if show_messages:
@@ -71,20 +75,15 @@ 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, speed=1.3)
+    audio_file = text_to_speech(response)
     return response, audio_file
 
-def text_to_speech(text, speed=1.3):
+def text_to_speech(text):
     tts = gTTS(text=text, lang='es')
     audio_fp = io.BytesIO()
     tts.write_to_fp(audio_fp)
     audio_fp.seek(0)
-    audio = AudioSegment.from_mp3(audio_fp)
-    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
+    return audio_fp
 
 def audio_play(audio_file):
     if audio_file is not None:
@@ -103,10 +102,10 @@ def main():
 
     audio_bytes = audiorecorder("Hablar ▶️", "Detener 🛑")   
     if audio_bytes:
-        audio_bytes = io.BytesIO(audio_bytes)
-        audio_bytes.seek(0)
-        audio_text = recognize_speech(audio_bytes)
-
+        audio_data = io.BytesIO(audio_bytes)  
+        audio_data.seek(0)  
+        audio_text = recognize_speech(audio_data)
+        
         if audio_text:
             output, audio_file = generate(audio_text, history=st.session_state.history)  
             display_recognition_result(audio_text, output, audio_file)