Voice Impersonation Detector API Design

Our project is a voice impersonation detector that determines whether an audio recording is real or AI-generated by analyzing speech patterns and anomalies.

🚀 API Overview

Endpoint: /mp3/{userid}

Instead of multiple endpoints, everything happens automatically after calling the client function /mp3. This calls a lambda function to upload the mp3 file, and this triggers the other backend functionsin a chain, so the user just uploads a file and waits for the results.

🛠️ Project Workflow

Step	AWS Service / Python Module Used	Lambda Function	Description	How is it Invoked?
Upload MP3	API Gateway & S3	final_proj_upload	Stores MP3 in S3 and starts processing	Invoked via /mp3/{userid} POST request
Speech-to-Text	Amazon Transcribe	final_proj_transcribe	Converts audio to text	Invoked when an MP3 is uploaded to S3
Sentiment Analysis	Amazon Comprehend	final_proj_analyze	Checks for tone inconsistencies	Invoked by final_proj_transcribe after transcription
Metadata Extraction	mutagen	final_proj_extract	Checks for suspicious patterns in metadata	Invoked by final_proj_analyze after sentiment analysis

📊 Diagram

🔄 API Workflow Documentation

1. Upload Function (final_proj_upload)

• Endpoint: /mp3/{userid}
• HTTP Verb: POST
• Request Parameters:
  • userid (int): ID of the user uploading the file
• Response:
  • Success
    • Status code: 200
    • Body: jobid
  • Bad Request
    • Status code: 400
    • Body: "no such user ..."
  • Internal Server Error
    • Status code: 500
    • Body: str(err)

2. Transcription Function (final_proj_transcribe)

• No direct HTTP endpoint
  • Triggered automatically via an S3 Event Notification when a new MP3 file is uploaded
• Request Parameters:
  • S3 key for the MP3 file
• Response:
  • Internal Server Error
    • Status code: 500
    • Body: str(err)

3. Analysis Function (final_proj_analyze)

• No direct HTTP Endpoint
  • Triggered by the transcription function via the Lambda SDK
• Request Parameters:
  • "File_key": "filename.txt"
  • "Text": "transcribed text..."
  • "Audio_key": "audio.mp3"
• Response:
  • Success
    • Status code: 200
    • Body:

      {
        "file_key": "file_key",
        "results_file": "bucketkey_results_file",
        "sentiment": "sentiment",
        "sentiment_scores": "sentiment_scores",
        "tone_score": "tone_score",
        "ai_speech_likelihood": "ai_likelihood",
        "s3_file": "s3://BUCKET_NAME/bucketkey_results_file"
      }

  • Error
    • Status code: 500
    • Body: str(err)

4. Extraction Function (final_proj_extract)

• No direct HTTP Endpoint
  • Triggered by the Analysis function via the Lambda SDK
• Request Parameters:
  • "File_key": "filename.txt"
  • "Text": "transcribed text..."
  • "Audio_key": "audio.mp3"
• Response:
  • Success
    • Status code: 200
    • Body:

      {
        "message": "Analysis completed and results file updated successfully",
        "bucketkey": "bucketkey_results_file"
      }

  • Error
    • Status code: 500
    • Body: str(err)

🗃️ Database Schema

Jobs Table

Column	Type	Description
jobid	int	Unique job ID
userid	int	User ID (links to the users table)
status	varchar	Job status (completed, error)
originaldatafile	varchar	Name of the uploaded MP3 file
datafilekey	varchar	S3 key for the MP3 file
resultsfilekey	varchar	S3 key for the final results file

Users Table

Column	Type	Description
userid	int	Unique user ID
username	varchar	User's name
pswhash	varchar	Hashed password for authentication