Create app.py

Gradio application for learning Arabic according to Noorani Qaida

app.py

@@ -0,0 +1,121 @@
+# Gradio application for learning Noorani Qaida
+
+p = pipeline("automatic-speech-recognition", model="jonatasgrosman/wav2vec2-large-xlsr-53-arabic")
+
+# Read excel file and store it in a dictionary
+def read_excel_data(file_path, sheet_name):
+    data_dict = {}
+    try:
+        # Read the Excel file
+        df = pd.read_excel(file_path, sheet_name=sheet_name)
+        # Iterate over the rows
+        for index, row in df.iterrows():
+            # Extract values from the row
+            key = row['correct']
+            value1 = row['close']
+            value2 = row['wrong']
+            # Store values in the dictionary
+            data_dict[key] = (value1, value2)
+        return data_dict
+    
+    except Exception as e:
+        print(f"Error reading Excel file: {e}")
+        
+
+# List of sample texts to read
+excel_file_path = "dataset/ASR_live_test.xlsx"
+sheet_name= 'sample_test'
+choices = read_excel_data(excel_file_path, sheet_name)
+
+def similar(a, b):
+    return SequenceMatcher(None, a, b).ratio()
+
+def transcribe(audio, reference_text):
+    time.sleep(1)
+    text = p(audio)["text"]
+    text = buckwalter.untrans(text)
+    state = text
+    if state is not None:
+        if similar(reference_text, state) > 0.75:
+            score = "great correct !"
+        elif similar(reference_text, state) > 0.50 and similar(reference_text, state) < 0.75:
+            score = "close"
+        else:
+            score = "wrong try again"
+        return state, score
+    else : 
+        print(" Null Object")
+
+# Gradio apps
+# get the list of the correct text to read 
+dropdown_choices = choices.keys()
+
+correct_callback = gr.CSVLogger()
+close_callback = gr.CSVLogger()
+wrong_callback = gr.CSVLogger()
+
+
+demo = gr.Blocks(theme=gr.themes.Soft())
+
+with demo :
+
+    gr.Markdown(
+        """
+    # Noorani Qaida Test Interface
+    Select a text from the list, read it aloud, and get the transcription along with the similarity score.
+    """
+    )
+
+    with gr.Row():
+        # Columns for inputs
+        with gr.Column():
+            correct_text = gr.Dropdown(label="Select Text to Read (Correct Text)", choices=dropdown_choices)
+            close_text = gr.Textbox(label="Close Text")
+            wrong_text = gr.Textbox(label="Wrong Text")
+
+            # Function to get the corresponding values of close and wrong text for each correct text
+            def get_text(correct_text):
+                close_text = gr.Textbox(label="Close Text", value=choices[correct_text][0])
+                wrong_text = gr.Textbox(label="Wrong Text", value=choices[correct_text][1])
+                return close_text, wrong_text
+
+            # Update the close and wrong text according to the selected correct text
+            correct_text.input(get_text, correct_text, [close_text, wrong_text])
+
+            audio_record = gr.Audio(type="filepath") 
+            
+            transcribe_correct_bn = gr.Button("Transcribe and Compare with Correct", scale=3)#remove scale as not effect
+            transcribe_close_bn = gr.Button("Transcribe and Compare with Close", scale=3)
+            transcribe_wrong_bn = gr.Button("Transcribe and Compare with Wrong", scale=3)
+            
+            with gr.Row():
+                clear_btn = gr.Button(value="Clear")
+
+        # Columns for outputs
+        with gr.Column():
+            trans_text = gr.Textbox(label="Transcription")
+            score = gr.Textbox(label="Score")
+            with gr.Row():
+                correct_flag_btn = gr.Button("Correct Flag", size="sm")
+                close_flag_btn = gr.Button("Close Flag", size="sm")
+                wrong_flag_btn = gr.Button("Wrong Flag", size="sm")
+    
+    # Setup the components to flag
+    correct_callback.setup([correct_text, trans_text, score, audio_record], "correct_flagged_data")
+    close_callback.setup([close_text, trans_text, score, audio_record], "close_flagged_data")
+    wrong_callback.setup([wrong_text, trans_text, score, audio_record], "wrong_flagged_data")
+
+    transcribe_correct_bn.click(fn=transcribe, inputs=[audio_record, correct_text], outputs=[trans_text, score])
+    transcribe_close_bn.click(fn=transcribe, inputs=[audio_record, close_text], outputs=[trans_text, score])
+    transcribe_wrong_bn.click(fn=transcribe, inputs=[audio_record, wrong_text], outputs=[trans_text, score])
+
+    # We can choose which components to flag 
+    correct_flag_btn.click(lambda *args: correct_callback.flag(args), [correct_text, trans_text, score, audio_record], None, preprocess=False)
+    close_flag_btn.click(lambda *args: close_callback.flag(args), [close_text, trans_text, score, audio_record], None, preprocess=False)
+    wrong_flag_btn.click(lambda *args: wrong_callback.flag(args), [wrong_text, trans_text, score, audio_record], None, preprocess=False)
+           
+    # Clear the recording
+    clear_btn.click(lambda: None, None, audio_record, queue=False)
+
+if __name__ == "__main__":
+    demo. Launch()