app.py

import gradio as gr
import os
import io
import json
from pydub import AudioSegment
from presidio_analyzer import AnalyzerEngine, PatternRecognizer, Pattern
from presidio_anonymizer import AnonymizerEngine
from google.cloud import speech

# ✅ Step 1: Set Google Cloud Credentials (File Uploaded in Hugging Face)
os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = "service-key.json"

# ✅ Step 2: Initialize Google Speech-to-Text API Client
client = speech.SpeechClient()

# ✅ Step 3: Initialize Presidio Analyzer & Anonymizer
analyzer = AnalyzerEngine()
anonymizer = AnonymizerEngine()

# ✅ Step 4: Define a Custom NHS Number Recognizer (UK Format)
nhs_pattern = Pattern(name="nhs_number_pattern", regex=r"\b\d{3}[- ]?\d{3}[- ]?\d{4}\b", score=0.85)
nhs_recognizer = PatternRecognizer(supported_entity="NHS_NUMBER", patterns=[nhs_pattern])

# Add NHS Recognizer to Presidio
analyzer.registry.add_recognizer(nhs_recognizer)

def transcribe_audio(audio_file):
    """
    Converts uploaded audio to text using Google Cloud Speech-to-Text.
    """
    # Convert audio to FLAC format for Google API compatibility
    audio = AudioSegment.from_file(audio_file)
    audio = audio.set_channels(1).set_frame_rate(16000)  # Ensure proper format
    
    # Save as FLAC in memory
    buffer = io.BytesIO()
    audio.export(buffer, format="flac")
    buffer.seek(0)

    # Send audio for transcription
    audio_recognition = speech.RecognitionAudio(content=buffer.read())
    config = speech.RecognitionConfig(language_code="en-GB")

    response = client.recognize(config=config, audio=audio_recognition)

    # Get transcribed text
    transcribed_text = " ".join([result.alternatives[0].transcript for result in response.results])
    
    return transcribed_text if transcribed_text else "No speech detected."

def redact_pii(audio_file):
    """
    1. Transcribes the audio using Google Speech-to-Text.
    2. Uses Presidio to redact sensitive PII from the transcript.
    """
    transcribed_text = transcribe_audio(audio_file)

    # Analyze and redact PII
    results = analyzer.analyze(
        text=transcribed_text,
        entities=["PERSON", "PHONE_NUMBER", "DATE_TIME", "LOCATION", "ID_NUMBER", "NHS_NUMBER"],
        language="en"
    )
    anonymized_text = anonymizer.anonymize(text=transcribed_text, analyzer_results=results)

    return transcribed_text, anonymized_text.text

# ✅ Step 5: Gradio UI for Audio Upload & Redaction
sample_transcript = """Example script for the demo:
'Hello, my name is Sarah Johnson. My NHS number is 123-456-7890, and I have an appointment with Dr. Smith on March 15th.
Please update my contact number to 07911 123456.'"""

iface = gr.Interface(
    fn=redact_pii,
    inputs=gr.Audio(type="filepath"),
    outputs=["text", "text"],
    title="Healthcare Call Redaction Demo",
    description=f"Upload a healthcare call recording, and Presidio will transcribe and anonymize sensitive data.\n\n"
                f"🔊 **Sample Script for Testing:**\n{sample_transcript}",
    examples=["sample_healthcare_call.wav"]
)


# ✅ Step 6: Launch the Gradio Web App
if __name__ == "__main__":
    iface.launch()