Update app.py

app.py

@@ -3,9 +3,9 @@ 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
-from audiorecorder import audiorecorder
 
 pre_prompt_text = "eres una IA conductual, tus respuestas serán breves."
 
@@ -17,31 +17,26 @@ 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)
 
+    with audio_recording as source:
+        audio = recognizer.record(source)
     try:
-        audio_data.seek(0)
-        audio = gTTS(audio_data.read(), lang='es')
-        audio_recording = sr.AudioFile(io.BytesIO(audio_data.read()))
-
-        with audio_recording as source:
-            audio = recognizer.record(source)
-
         audio_text = recognizer.recognize_google(audio, language="es-ES")
         if show_messages:
             st.subheader("Texto Reconocido:")
             st.write(audio_text)
-            st.success("Reconocimiento completado.")
+            st.success("Reconocimiento de voz completado.")
     except sr.UnknownValueError:
-        st.warning("¡Habla fuerte y claro!")
+        st.warning("No se pudo reconocer el audio. ¿Intentaste grabar algo?")
         audio_text = ""
     except sr.RequestError:
-        st.error("¡Háblame para comenzar!")
+        st.error("Háblame para comenzar!")
         audio_text = ""
 
     return audio_text
 
 def format_prompt(message, history):
-    preprompt = "Este es el inicio de la conversación. "
     prompt = "<s>"
 
     for user_prompt, bot_response in history:
@@ -49,9 +44,9 @@ def format_prompt(message, history):
         prompt += f" {bot_response}</s> "
 
     prompt += f"[INST] {message} [/INST]"
-    return preprompt + prompt
+    return prompt
 
-def generate(audio_text, history, temperature=None, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0):
+def generate(audio_text, history, temperature=None, max_new_tokens=512, 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
@@ -74,9 +69,9 @@ def generate(audio_text, history, temperature=None, max_new_tokens=256, top_p=0.
 
     for response_token in stream:
         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, speed=1.3):
@@ -84,41 +79,39 @@ def text_to_speech(text, speed=1.3):
     audio_fp = io.BytesIO()
     tts.write_to_fp(audio_fp)
     audio_fp.seek(0)
-    audio = AudioSegment.from_file(audio_fp, format="mp3")
+    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
 
-def play_audio(audio_file):
-    audio_data = audio_file.read()
-    b64_audio = base64.b64encode(audio_data).decode("utf-8")
-    st.audio(f"data:audio/mp3;base64,{b64_audio}", format="audio/mp3", start_time=0)
+def audio_play(audio_fp):
+    st.audio(audio_fp.read(), format="audio/mp3", start_time=0)
 
-def main():
-    st.title("Chatbot de Voz a Voz")
-    audio_text = ""
+def display_recognition_result(audio_text, output, audio_file):
+    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)
+
+def main():
     if not st.session_state.pre_prompt_sent:
         st.session_state.pre_prompt_sent = True
-        st.session_state.history.append((pre_prompt_text, ""))
-        pre_prompt_audio_file = text_to_speech(pre_prompt_text)
-        play_audio(pre_prompt_audio_file)
-
-    audio_data = audiorecorder.record("Hablar ▶️", "Detener 🛑")
 
-    if not audio_data.empty():
-        st.audio(audio_data.export().read(), format="audio/mp3")
+    audio_bytes = audiorecorder("Hablar ▶️", "Detener 🛑")   
+    if audio_bytes:
+        st.audio(audio_bytes, format="audio/wav")
+        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)
-
-            if audio_text:
-                st.session_state.history.append((audio_text, output))
-            if audio_file is not None:
-                play_audio(audio_file)
+            output, audio_file = generate(audio_text, history=st.session_state.history)  
+            display_recognition_result(audio_text, output, audio_file)
 
 if __name__ == "__main__":
     main()