Update app.py

app.py

@@ -1,9 +1,12 @@
 import streamlit as st
+import base64
 import io
 from huggingface_hub import InferenceClient
 from gtts import gTTS
 from audiorecorder import audiorecorder
+import wave
 import speech_recognition as sr
+from pydub import AudioSegment
 
 pre_prompt_text = "eres una IA conductual, tus respuestas serán breves."
 
@@ -15,9 +18,7 @@ if "pre_prompt_sent" not in st.session_state:
 
 def recognize_speech(audio_data, show_messages=True):
     recognizer = sr.Recognizer()
-
-    audio_data.seek(0)  # Asegurémonos de que el puntero esté al principio
-    audio_recording = sr.AudioFile(io.BytesIO(audio_data.read()))
+    audio_recording = sr.AudioFile(audio_data)
 
     with audio_recording as source:
         audio = recognizer.record(source)
@@ -47,7 +48,7 @@ def format_prompt(message, history):
     prompt += f"[INST] {message} [/INST]"
     return prompt
 
-def generate(audio_text, history, temperature=None, max_new_tokens=512, top_p=0.95, repetition_penalty=1.0):
+def generate(audio_text, history, temperature=None, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0):
     client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
 
     temperature = float(temperature) if temperature is not None else 0.9
@@ -72,34 +73,45 @@ 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):
+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)
-    return audio_fp
+    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 audio_play(audio_file):
-    if audio_file is not None:
-        audio_binary = audio_file.read()
-        st.audio(audio_binary, format="audio/mp3", start_time=0)
+def audio_play(audio_fp):
+    st.audio(audio_fp.read(), format="audio/mp3", start_time=0)
 
 def display_recognition_result(audio_text, output, audio_file):
     if audio_text:
         st.session_state.history.append((audio_text, output))  
-    audio_play(audio_file)
+
+    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)
 
 def main():
     if not st.session_state.pre_prompt_sent:
         st.session_state.pre_prompt_sent = True
 
-    audio_bytes = audiorecorder("Hablar ▶️", "Detener 🛑")   
-    if audio_bytes:
-        audio_data = io.BytesIO(audio_bytes)  
-        audio_text = recognize_speech(audio_data)
+    audio_data = audiorecorder(start_prompt="Hablar ▶️", stop_prompt="Detener 🛑")
+    
+    if audio_data and 'bytes' in audio_data:
+        audio_bytes = audio_data['bytes']
+        st.audio(audio_bytes, format="audio/wav")
+        audio_data_io = io.BytesIO(audio_bytes)  
+        audio_data_io.seek(0)  
+        audio_text = recognize_speech(audio_data_io)
         
         if audio_text:
             output, audio_file = generate(audio_text, history=st.session_state.history)