Update app.py

app.py

@@ -1,39 +1,77 @@
 import gradio as gr
+from transformers import pipeline
+import numpy as np
 
+# Initialize the automatic speech recognition pipeline using a pre-trained model
+transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-base.en")
 
-def click_js():
-    return """function audioRecord() {
-    var xPathRes = document.evaluate ('//*[contains(@class, "record")]', document, null, XPathResult.FIRST_ORDERED_NODE_TYPE, null); 
-    xPathRes.singleNodeValue.click();}"""
+# Global variables to store the accumulated audio data and its streaming rate
+audio_data = None
+streaming_rate = None
 
+def capture_audio(stream, new_chunk):
+    """
+    Function to capture streaming audio and accumulate it in a global variable.
 
-def action(btn):
-    """Changes button text on click"""
-    if btn == 'Speak': return 'Stop'
-    else: return 'Speak'
+    Args:
+        stream (numpy.ndarray): The accumulated audio data up to this point.
+        new_chunk (tuple): A tuple containing the sampling rate and the new audio data chunk.
 
+    Returns:
+        numpy.ndarray: The updated stream with the new chunk appended.
+    """
+    global audio_data
+    global streaming_rate
 
-def check_btn(btn):
-    """Checks for correct button text before invoking transcribe()"""
-    if btn != 'Speak': raise Exception('Recording...')
+    # Extract sampling rate and audio chunk, normalize the audio
+    sr, y = new_chunk
+    streaming_rate = sr
+    y = y.astype(np.float32)
+    y /= np.max(np.abs(y))
 
+    # Concatenate new audio chunk to the existing stream or start a new one
+    if stream is not None:
+        stream = np.concatenate([stream, y])
+    else:
+        stream = y
 
-def transcribe():
-    return 'Success'
+    # Update the global variable with the new audio data
+    audio_data = stream
+    return stream
 
-with gr.Blocks() as demo:
-    msg = gr.Textbox()
-    audio_box = gr.Audio(label="Audio", sources="microphone", type="filepath", elem_id='audio')
+def get_transcript():
+    """
+    Function to transcribe the accumulated audio data.
 
-    with gr.Row():
-        audio_btn = gr.Button('Speak')
-        clear = gr.Button("Clear")
+    Returns:
+        str: The transcription of the accumulated audio data.
+    """
+    global audio_data
+    global streaming_rate
+
+    # Transcribe the audio data if available
+    if audio_data is not None and streaming_rate is not None:
+        transcript = transcriber({"sampling_rate": streaming_rate, "raw": audio_data})["text"]
+        return transcript
+    return ""
 
-    audio_btn.click(fn=action, inputs=audio_btn, outputs=audio_btn).\
-              then(fn=lambda: None, js=click_js()).\
-              then(fn=check_btn, inputs=audio_btn).\
-              success(fn=transcribe, outputs=msg)
+# Building the Gradio interface using Blocks
+with gr.Blocks() as demo:
+    with gr.Row():
+        with gr.Column():
+            # State variable to manage the streaming data
+            state = gr.State()
+            # Audio component for real-time audio capture from the microphone
+            audio = gr.Audio(sources=["microphone"], streaming=True, type="numpy")
+            # Textbox for displaying the transcription
+            transcript_box = gr.Textbox(label="Transcript")
+            # Button to initiate transcription of the captured audio
+            rfrsh_btn = gr.Button("Refresh")
 
-    clear.click(lambda: None, None, msg, queue=False)
+            # Streaming setup to handle real-time audio capture
+            audio.stream(fn=capture_audio, inputs=[state, audio], outputs=[state])
+            # Button click setup to trigger transcription
+            rfrsh_btn.click(fn=get_transcript, outputs=[transcript_box])
 
-demo.queue().launch(debug=True)
+# Launch the Gradio interface
+demo.launch()