Update app.py

app.py

@@ -1,48 +1,41 @@
-import io
+import streamlit as st
 import base64
+import io
+from huggingface_hub import InferenceClient
 from gtts import gTTS
-import streamlit as st
+from audiorecorder import audiorecorder
 import speech_recognition as sr
-from huggingface_hub import InferenceClient
-from streamlit_mic_recorder import mic_recorder
-import wave
-import numpy as np
+from pydub import AudioSegment
 import os
 
-pre_prompt_text = "eres una IA conductual, tus respuestas serán breves."
-temp_audio_file_path = "./output.wav"
-
 if "history" not in st.session_state:
     st.session_state.history = []
 
-if "pre_prompt_sent" not in st.session_state:
-    st.session_state.pre_prompt_sent = False
-
 def recognize_speech(audio_data, show_messages=True):
     recognizer = sr.Recognizer()
-
-    with io.BytesIO(audio_data) as audio_file:
-        try:
-            audio_text = recognizer.recognize_google(audio_file, language="es-ES")
-            if show_messages:
-                st.subheader("Texto Reconocido:")
-                st.write(audio_text)
-                st.success("Reconocimiento de voz completado.")
-        except sr.UnknownValueError:
-            st.warning("No se pudo reconocer el audio. ¿Intentaste grabar algo?")
-            audio_text = ""
-        except sr.RequestError:
-            st.error("Hablame para comenzar!")
-            audio_text = ""
+    audio_recording = sr.AudioFile(audio_data)
+
+    with audio_recording as source:
+        audio = recognizer.record(source)
+
+    try:
+        audio_text = recognizer.recognize_google(audio, language="es-ES")
+        if show_messages:
+            st.subheader("Texto Reconocido:")
+            st.write(audio_text)
+            st.success("Reconocimiento de voz completado.")
+    except sr.UnknownValueError:
+        st.warning("No se pudo reconocer el audio. ¿Intentaste grabar algo?")
+        audio_text = ""
+    except sr.RequestError:
+        st.error("Hablame para comenzar!")
+        audio_text = ""
 
     return audio_text
 
 def format_prompt(message, history):
     prompt = "<s>"
 
-    if not st.session_state.pre_prompt_sent:
-        prompt += f"[INST]{pre_prompt_text}[/INST]"
-
     for user_prompt, bot_response in history:
         prompt += f"[INST] {user_prompt} [/INST]"
         prompt += f" {bot_response}</s> "
@@ -64,7 +57,8 @@ def generate(audio_text, history, temperature=None, max_new_tokens=512, top_p=0.
         top_p=top_p,
         repetition_penalty=repetition_penalty,
         do_sample=True,
-        seed=42,)
+        seed=42,
+    )
 
     formatted_prompt = format_prompt(audio_text, history)
     stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
@@ -82,39 +76,36 @@ def text_to_speech(text, speed=1.3):
     audio_fp = io.BytesIO()
     tts.write_to_fp(audio_fp)
     audio_fp.seek(0)
-    return audio_fp
-
-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))  
-
-    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 = 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():
-    if not st.session_state.pre_prompt_sent:
-        st.session_state.pre_prompt_sent = True
-
-    audio = mic_recorder(start_prompt="▶️", stop_prompt="🛑", key='recorder')
-
-    if audio:       
-        st.audio(audio['bytes'])
-       
-        audio_bytes = audio["bytes"]
-        sample_width = audio["sample_width"]  # 2 bytes per sample for 16-bit PCM
-        sample_rate = audio["sample_rate"]  # 44.1 kHz sample rate
-        num_channels = 1  # 1 channel for mono, 2 for stereo
-    
-        with wave.open(temp_audio_file_path, 'w') as wave_file:
-            wave_file.setnchannels(num_channels)
-            wave_file.setsampwidth(sample_width)
-            wave_file.setframerate(sample_rate)
-            wave_file.writeframes(audio_bytes)    
+    st.title("Chatbot de Voz a Voz")
+    audio_text = ""
+    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 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()