Update app.py

app.py

@@ -1,45 +1,33 @@
-# voicepulse.py (multilingual feedback transcriber and word cloud generator with export)
 
 import gradio as gr
-import numpy as np
 from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
-import matplotlib.pyplot as plt
 from wordcloud import WordCloud
-import nltk
-import io
+import matplotlib.pyplot as plt
+import numpy as np
+import soundfile as sf
 import torch
+import io
 import csv
-from gtts import gTTS
-
-nltk.download("stopwords")
-stop_words = set(nltk.corpus.stopwords.words("english"))
+from datetime import datetime
+import os
 
-# Translation model for multilingual -> English
+# 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
 
-# In-memory feedback word list and archive
+# Persistent state
+feedback_records = []
 feedback_words = []
-all_feedback = []
-
-# Sample audios
-sample_texts = {
-    "Telugu Sample": "మీ సేవలు చాలా బాగున్నాయి. మేము చాలా సంతృప్తిగా ఉన్నాము.",
-    "Hindi Sample": "आपकी सेवा बहुत अच्छी थी और हम संतुष्ट हैं।",
-    "Tamil Sample": "உங்கள் சேவை மிகவும் சிறந்ததாக இருந்தது. நாங்கள் திருப்தி அடைந்தோம்.",
-    "English Sample": "Your support team was helpful and responsive."
-}
 
-def generate_sample_audio(text, lang_code):
-    tts = gTTS(text, lang=lang_code)
-    tts.save("sample_full.mp3")
-
-    from pydub import AudioSegment
-    full_audio = AudioSegment.from_mp3("sample_full.mp3")
-    short_audio = full_audio[:3000]  # first 3 seconds
-    short_audio.export("sample.mp3", format="mp3")
-    return "sample.mp3"
+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(
@@ -51,90 +39,76 @@ def translate(text, src_lang, tgt_lang="eng_Latn"):
         max_length=400,
         device=device
     )
-    result = translation_pipeline(text)
-    return result[0]['translation_text']
-
-def get_transcription(audio, language):
-    sr, y = audio
-    y = y.astype(np.float32)
-    y /= np.max(np.abs(y))
-
-    if language == "English":
-        transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-base.en")
-        return transcriber({"sampling_rate": sr, "raw": y})["text"]
-
-    model_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")
-    }
-
-    model_name, src_lang = model_map[language]
+    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)
-    text = transcriber({"sampling_rate": sr, "raw": y})["text"]
-    return translate(text, src_lang)
+    transcription = transcriber(audio_np)["text"]
+
+    if language != "English":
+        transcription = translate(transcription, src_lang=src_lang)
 
-def process_feedback(audio, language):
-    transcription = get_transcription(audio, language)
+    # Summarize
+    summary = summarizer(transcription, max_length=60, min_length=10, do_sample=False)[0]["summary_text"]
 
-    # Use summarization to extract core feedback idea
-    summarizer = pipeline("summarization", model="sshleifer/distilbart-cnn-12-6")
-    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
+    })
 
-    # Save for download
-    all_feedback.append({"Language": language, "Transcription": transcription, "Summary": summary})
+    # Word cloud
+    feedback_words.extend(summary.lower().split())
+    wordcloud_img = generate_wordcloud(feedback_words)
 
-    # Extract meaningful words from summary
-    words = [w for w in summary.lower().split() if w.isalpha() and w not in stop_words]
-    feedback_words.extend(words)
+    return summary, wordcloud_img
 
-    freq = {w: feedback_words.count(w) for w in set(feedback_words)}
-    wc = WordCloud(width=800, height=400, background_color="white").generate_from_frequencies(freq)
+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.imshow(wc, interpolation="bilinear")
-    plt.axis("off")
     plt.savefig(buf, format="png")
+    plt.close(fig)
     buf.seek(0)
-    image = plt.imread(buf, format="png")
-
-    return summary, image
+    return buf
 
 def export_to_csv():
-    with open("feedback_export.csv", "w", newline="") as csvfile:
-        fieldnames = ["Language", "Transcription", "Summary"]
-        writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
+    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 all_feedback:
+        for row in feedback_records:
             writer.writerow(row)
-    return "feedback_export.csv"
-
+    return filename
 
 demo = gr.Blocks()
 
 with demo:
     gr.Markdown("# 🎙️ VoicePulse Multilingual Feedback Collector")
-    gr.Markdown("""
-    🗣️ **VoicePulse** lets you speak feedback in your language — Telugu, Hindi, Tamil, Kannada, or English.
-    It transcribes, translates, and summarizes the feedback, building a live word cloud to show what people care about.
-    Try speaking or use sample audio buttons below!
-    """)
+    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():
-        audio_input = gr.Audio(type="numpy", label="🎤 Speak your feedback")
-        lang_dropdown = gr.Dropdown(label="🌐 Language", choices=["English", "Hindi", "Telugu", "Tamil", "Kannada"], value="English")
+        gr.Markdown("🔴 **Select language of the uploaded file**")
 
     with gr.Row():
-        submit_btn = gr.Button("Process Feedback")
-        
+        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():
-        gr.Markdown("### 🎧 Sample Feedback (Telugu, Hindi, Tamil, English)")
-        sample_btn_te = gr.Button("🔉 Telugu Sample")
-        sample_btn_hi = gr.Button("🔉 Hindi Sample")
-        sample_btn_ta = gr.Button("🔉 Tamil Sample")
-        sample_btn_en = gr.Button("🔉 English Sample")
-        sample_audio = gr.Audio(label="🔊 Sample Audio Output (Preview)")
-        
+        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")
@@ -145,9 +119,5 @@ with demo:
 
     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)
-    sample_btn_te.click(lambda: generate_sample_audio(sample_texts["Telugu Sample"], 'te'), inputs=[], outputs=sample_audio)
-    sample_btn_hi.click(lambda: generate_sample_audio(sample_texts["Hindi Sample"], 'hi'), inputs=[], outputs=sample_audio)
-    sample_btn_ta.click(lambda: generate_sample_audio(sample_texts["Tamil Sample"], 'ta'), inputs=[], outputs=sample_audio)
-    sample_btn_en.click(lambda: generate_sample_audio(sample_texts["English Sample"], 'en'), inputs=[], outputs=sample_audio)
 
 demo.launch()