M12faiez/Audio_Based_Stock_Analysis_Agent_PoC

1import pandas as pd2import gradio as gr3from transformers import pipeline4import whisper5 6 7# Load the TAPAS pipeline for table-question-answering8pipe = pipeline("table-question-answering", model="google/tapas-base-finetuned-wtq")9 10 11# Function to load the CSV file and prepare the table12def load_csv_table(csv_file):13    """14    Load the CSV file and convert it into a Pandas DataFrame.15    """16    df = pd.read_csv(csv_file.name)  # Gradio provides the file as an object with a .name attribute17    table = df.astype(str)  # Convert all values to strings for compatibility18    return table19 20 21# Function to use TAPAS pipeline to answer a question22def answer_question_with_pipeline(table, question):23    """24    Use TAPAS pipeline to answer the question based on the table.25    """26    result = pipe(table=table.to_dict(orient="records"), query=question)27    return result["answer"]28 29 30# Load the Whisper model for audio transcription31whisper_model = whisper.load_model("base")32 33def transcribe(audio):34    """35    Transcribe audio using the Whisper model.36    """37    # Load and preprocess audio38    audio = whisper.load_audio(audio)39    audio = whisper.pad_or_trim(audio)40 41    # Generate log-Mel spectrogram and transcribe42    mel = whisper.log_mel_spectrogram(audio).to(whisper_model.device)43    _, probs = whisper_model.detect_language(mel)44    print(f"Detected language: {max(probs, key=probs.get)}")45 46    # Decode the audio and return text47    options = whisper.DecodingOptions()48    result = whisper.decode(whisper_model, mel, options)49    return result.text50 51 52def process(csv_file, audio):53    """54    Process the uploaded CSV file and audio to generate an answer.55    """56    # Step 1: Transcribe the audio to get the question57    question = transcribe(audio)58 59    # Step 2: Load the table from the uploaded CSV file60    table = load_csv_table(csv_file)61 62    # Step 3: Use TAPAS to answer the transcribed question63    answer = answer_question_with_pipeline(table, question)64 65    return question, answer66 67 68iface = gr.Interface(69    fn=process,  # Main function integrating Whisper and TAPAS70    inputs=[71        gr.File(label="Upload CSV File and ask quuestion i.e which company_name has most volume value"),  # Upload CSV file72        gr.Audio(type="filepath", label="Record or Upload Audio for Question (wait for audio to be processed if getting error, click submit again)")  # Record/upload audio73    ],74    outputs=[75        gr.Textbox(label="Transcribed Question"),  # Display transcribed question76        gr.Textbox(label="Answer from Table")  # Display TAPAS answer77    ],78    title="Table Question Answering with Audio Input",79    description="Upload a CSV file, ask your question via audio, and get answers using TAPAS."80)81 82# Launch Gradio Interface83iface.launch()84