Update app.py

app.py

@@ -1,19 +1,16 @@
 import streamlit as st
-import torch
-import soundfile as sf
 import pyttsx3
 import threading
-from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+import wave
+import io
+import speech_recognition as sr
 from gradio_client import Client
+import streamlit.components.v1 as components
 
 # Initialize session state
 if "messages" not in st.session_state:
     st.session_state["messages"] = []  # Store chat history
 
-# Load the Wav2Vec 2.0 model and processor from Hugging Face
-processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-large-960h")
-model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-960h")
-
 # Function to generate a response using Gradio client
 def generate_response(query):
     try:
@@ -52,34 +49,69 @@ def update_chat_history():
         if "bot" in msg:
             st.markdown(f"<div class='chat-bubble bot-message'><strong>Bot:</strong> {msg['bot']}</div>", unsafe_allow_html=True)
 
-# Function to recognize speech using Hugging Face's Wav2Vec 2.0
-def recognize_speech_huggingface():
-    st.info("Listening... Speak into the microphone.")
-    fs = 16000  # Sample rate in Hz
-    duration = 5  # Duration in seconds
-
-    # Record the audio using sounddevice or use a pre-recorded file
-    # (Here we're using soundfile to record from microphone)
-    audio_data = sd.rec(int(duration * fs), samplerate=fs, channels=1, dtype='int16')
-    sd.wait()
-    # Save the audio file to a temporary buffer
-    sf.write('audio.wav', audio_data, fs)
-
-    # Read the audio file using soundfile and process it
-    audio_input, _ = sf.read('audio.wav')
+# Function to recognize speech from audio received as bytes
+def recognize_speech_from_audio(audio_bytes):
+    st.info("Processing audio...")
 
-    # Preprocess the audio and recognize the speech
-    inputs = processor(audio_input, return_tensors="pt", sampling_rate=fs)
-    with torch.no_grad():
-        logits = model(input_values=inputs.input_values).logits
+    # Convert byte stream to audio file
+    audio_data = io.BytesIO(audio_bytes)
+    recognizer = sr.Recognizer()
 
-    # Decode the logits to text
-    predicted_ids = torch.argmax(logits, dim=-1)
-    recognized_text = processor.decode(predicted_ids[0])
+    # Recognize speech from the audio data
+    with sr.AudioFile(audio_data) as source:
+        audio = recognizer.record(source)
 
-    st.session_state["user_input"] = recognized_text
-    st.success(f"Recognized Text: {recognized_text}")
-    handle_user_input(recognized_text)
+    try:
+        recognized_text = recognizer.recognize_google(audio)
+        st.session_state["user_input"] = recognized_text
+        st.success(f"Recognized Text: {recognized_text}")
+        handle_user_input(recognized_text)
+    except sr.UnknownValueError:
+        st.error("Sorry, I couldn't understand the audio.")
+    except sr.RequestError:
+        st.error("Could not request results; please check your internet connection.")
+
+# JavaScript for audio recording and sending data to Streamlit
+audio_recorder_html = """
+    <script>
+        let audioChunks = [];
+        let mediaRecorder;
+        
+        function startRecording() {
+            navigator.mediaDevices.getUserMedia({ audio: true })
+                .then(function(stream) {
+                    mediaRecorder = new MediaRecorder(stream);
+                    mediaRecorder.ondataavailable = function(event) {
+                        audioChunks.push(event.data);
+                    };
+                    mediaRecorder.onstop = function() {
+                        const audioBlob = new Blob(audioChunks, { type: 'audio/wav' });
+                        const reader = new FileReader();
+                        reader.onloadend = function() {
+                            const audioBase64 = reader.result.split(',')[1];
+                            window.parent.postMessage({ 'type': 'audio_data', 'audio': audioBase64 }, '*');
+                        };
+                        reader.readAsDataURL(audioBlob);
+                    };
+                    mediaRecorder.start();
+                });
+        }
+        
+        function stopRecording() {
+            mediaRecorder.stop();
+        }
+        
+        function handleStartStop() {
+            if (mediaRecorder && mediaRecorder.state === "recording") {
+                stopRecording();
+            } else {
+                startRecording();
+            }
+        }
+    </script>
+    <button onclick="handleStartStop()">Start/Stop Recording</button>
+    <p>Click the button to start/stop audio recording.</p>
+"""
 
 # Main Streamlit app
 st.set_page_config(page_title="Llama2 Chatbot", page_icon="🤖", layout="wide")
@@ -138,7 +170,7 @@ st.markdown(
     """
     Welcome to the *Llama2 Chatbot*!   
     - *Type* your message below, or   
-    - *Use the microphone* to speak to the bot.   
+    - *Speak* to the bot using your microphone.   
     """
 )
 
@@ -154,10 +186,17 @@ with chat_history_container:
         if submit_button:
             handle_user_input(user_input)
 
-    # Separate button for speech recognition outside of the form
-    if st.button("Speak"):
-        recognize_speech_huggingface()
+    # Display JavaScript for audio recording
+    components.html(audio_recorder_html, height=300)
 
-    st.markdown("### Chat History")
     # Update chat history on every interaction
-    update_chat_history()
+    update_chat_history()
+
+# Listening to the audio data sent by JavaScript
+def process_audio_data():
+    audio_data = st.experimental_get_query_params().get('audio', [None])[0]
+    if audio_data:
+        recognize_speech_from_audio(audio_data)
+
+# Call the function to process audio if available
+process_audio_data()