Update app.py

app.py

@@ -1,11 +1,12 @@
 import streamlit as st
 import base64
 import io
-import audiorecorder
-import speech_recognition as sr
 from huggingface_hub import InferenceClient
+from gtts import gTTS
+from audiorecorder import audiorecorder
+import speech_recognition as sr
 
-pre_prompt_text = "You are a behavioral AI, your answers should be brief, stoic and humanistic."
+pre_prompt_text = ""
 
 if "history" not in st.session_state:
     st.session_state.history = []
@@ -49,7 +50,7 @@ def format_prompt(message, history):
     prompt += f"[INST] {message} [/INST]"
     return prompt
 
-def generate(audio_text, history, temperature=None, max_new_tokens=2048, 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
@@ -67,19 +68,12 @@ def generate(audio_text, history, temperature=None, max_new_tokens=2048, top_p=0
     formatted_prompt = format_prompt(audio_text, history)
     stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
     response = ""
-    response_tokens = []
-    total_tokens = 0
-
-    progress_bar = st.progress(0)
 
     for response_token in stream:
-        total_tokens += len(response_token.token.text)
-        response_tokens.append(response_token.token.text)
-        response = ' '.join(response_tokens).replace('</s>', '')
-        progress = min(total_tokens / max_new_tokens, 1.0)  # Asegurar que el progreso esté en el rango [0.0, 1.0]
-        progress_bar.progress(progress)
-
-    audio_file = text_to_speech(response)
+        response += response_token.token.text
+    
+    response = ' '.join(response.split()).replace('</s>', '')
+    audio_file = text_to_speech(response, speed=1.3)
     return response, audio_file
 
 def text_to_speech(text):
@@ -90,28 +84,46 @@ def text_to_speech(text):
     return audio_fp
 
 def main():
-    option = st.radio("Select Input Method:", ("Text", "Voice"))
-
-    if option == "Text":
-        prompt = st.text_area("Enter your prompt here:")
-    else:
-        st.write("Push and hold the button to record.")
-        audio_data = audiorecorder.audiorecorder("Push to Talk", "Stop Recording...")
-
-        if not audio_data.empty():
-            st.audio(audio_data.export().read(), format="audio/wav")
-            audio_data.export("audio.wav", format="wav")
-            prompt = recognize_speech("audio.wav")
-            st.text("Recognized prompt:")
-            st.write(prompt)
-
-    if prompt:
-        output, audio_file = generate(prompt, history=st.session_state.history)  
-
-        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)
+    audio_data = audiorecorder("Push to Talk", "Stop Recording...")
+
+    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 audio_text:
+            output, audio_file = generate(audio_text, history=st.session_state.history)  
+
+            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)
+
+# Voice Activity Detection (VAD) using speech_recognition
+def vad_loop():
+    command = ""
+    while True:
+        with sr.Microphone() as source:
+            st.write("Listening...")
+            recognizer = sr.Recognizer()
+            recognizer.adjust_for_ambient_noise(source)
+            audio = recognizer.listen(source)
+
+            try:
+                command = recognizer.recognize_google(audio, language="en-US")
+                st.write(f"Command: {command}")
+            except sr.UnknownValueError:
+                st.write("Could not understand audio")
+            except sr.RequestError as e:
+                st.write(f"Error: {e}")
+
+        if command.lower() == "xaman":
+            st.write("Voice capture activated. Say 'Detente', 'Alto', or 'Basta' to stop.")
+            audio_data = audiorecorder("Push to Talk", "Stop Recording...")
+            main()
+        elif command.lower() in ["detente", "alto", "basta"]:
+            st.write("Voice capture stopped.")
+            break
 
 if __name__ == "__main__":
-    main()
+    vad_loop()