app.py

import gradio as gr
from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
from wordcloud import WordCloud
import matplotlib.pyplot as plt
import numpy as np
import soundfile as sf
import torch
import io
import csv
from datetime import datetime
import os

# Set up translation and summarization models
translation_model = AutoModelForSeq2SeqLM.from_pretrained("facebook/nllb-200-distilled-600M")
translation_tokenizer = AutoTokenizer.from_pretrained("facebook/nllb-200-distilled-600M")
summarizer = pipeline("summarization", model="sshleifer/distilbart-cnn-12-6")
device = 0 if torch.cuda.is_available() else -1

# Persistent state
feedback_records = []
feedback_words = []

lang_map = {
    "Hindi": ("theainerd/Wav2Vec2-large-xlsr-hindi", "hin_Deva"),
    "Telugu": ("anuragshas/wav2vec2-large-xlsr-53-telugu", "tel_Telu"),
    "Tamil": ("Harveenchadha/vakyansh-wav2vec2-tamil-tam-250", "tam_Taml"),
    "Kannada": ("vasista22/whisper-kannada-medium", "kan_Knda"),
    "English": ("openai/whisper-base.en", None)
}

def translate(text, src_lang, tgt_lang="eng_Latn"):
    translation_pipeline = pipeline(
        "translation",
        model=translation_model,
        tokenizer=translation_tokenizer,
        src_lang=src_lang,
        tgt_lang=tgt_lang,
        max_length=400,
        device=device
    )
    return translation_pipeline(text)[0]["translation_text"]

def process_feedback(audio_np, language):
    if audio_np is None:
        return "No audio provided", None

    # Transcribe
    model_name, src_lang = lang_map[language]
    transcriber = pipeline("automatic-speech-recognition", model=model_name)
    transcription = transcriber(audio_np)["text"]

    if language != "English":
        transcription = translate(transcription, src_lang=src_lang)

    # Summarize
    summary = summarizer(transcription, max_length=60, min_length=10, do_sample=False)[0]["summary_text"]

    # Save record
    feedback_records.append({
        "timestamp": datetime.utcnow().isoformat(),
        "language": language,
        "transcription": transcription,
        "summary": summary
    })

    # Word cloud
    feedback_words.extend(summary.lower().split())
    wordcloud_img = generate_wordcloud(feedback_words)

    return summary, wordcloud_img

def generate_wordcloud(words):
    wordcloud = WordCloud(width=600, height=300, background_color="white").generate(" ".join(words))
    fig, ax = plt.subplots(figsize=(6, 3))
    ax.imshow(wordcloud, interpolation="bilinear")
    ax.axis("off")
    buf = io.BytesIO()
    plt.savefig(buf, format="png")
    plt.close(fig)
    buf.seek(0)
    return buf

def export_to_csv():
    filename = "feedback_export.csv"
    with open(filename, mode='w', newline='', encoding='utf-8') as file:
        writer = csv.DictWriter(file, fieldnames=["timestamp", "language", "transcription", "summary"])
        writer.writeheader()
        for row in feedback_records:
            writer.writerow(row)
    return filename

demo = gr.Blocks()

with demo:
    gr.Markdown("# 🎙️ VoicePulse Multilingual Feedback Collector")
    gr.Markdown(
        "🗣️ **VoicePulse** lets you upload or speak feedback in your language — Telugu, Hindi, Tamil, Kannada, or English.\n"
        "It transcribes, translates, and summarizes the feedback, building a live word cloud to show what people care about."
    )

    with gr.Row():
        gr.Markdown("🔴 **Select language of the uploaded file**")

    with gr.Row():
        audio_input = gr.Audio(type="numpy", label="📤 Upload your feedback audio")
        lang_dropdown = gr.Dropdown(label="🌐 Language", choices=["English", "Hindi", "Telugu", "Tamil", "Kannada"], value="English")

    with gr.Row():
        submit_btn = gr.Button("✅ Process Feedback")

    with gr.Row():
        summary_out = gr.Textbox(label="📝 Summarized Feedback")
        wordcloud_out = gr.Image(type="pil", label="☁️ Word Cloud of All Feedback")

    with gr.Row():
        export_btn = gr.Button("📁 Export Feedback to CSV")
        csv_file_output = gr.File(label="📄 Download CSV")

    submit_btn.click(process_feedback, inputs=[audio_input, lang_dropdown], outputs=[summary_out, wordcloud_out])
    export_btn.click(export_to_csv, inputs=[], outputs=csv_file_output)

demo.launch()