Skip to main content
📦 GitHub: Dataspike Deepfake SDK for Android
The Dataspike Deepfake SDK for Android provides a compact, high-performance framework for detecting AI-generated or manipulated video and audio directly on-device — no internet connection required. Built on Dataspike’s deepfake detection engine, the SDK delivers real-time analysis, low latency, and enterprise-grade accuracy, even on mid-tier Android devices. It supports both video and streaming audio detection and integrates seamlessly with Jetpack Compose, WebRTC, and standard Android views.
⚡ Key advantage: All inference runs locally on the device using a lightweight yet powerful model — ensuring privacy, speed, and reliability in fully offline scenarios.
Github

​
Highlights

  • 🚀 On‑device inference — no internet dependency, no data upload.
  • 🧠 Optimized neural model — compact yet accurate, built for real‑time response.
  • 🎥 Video & streaming support — compatible with SurfaceView, TextureView, and WebRTC VideoTrack.
  • 🔒 Privacy-first architecture — no cloud processing; user content never leaves the device.
  • ⚙️ Kotlin‑first API — clean, composable, and easy to integrate.
  • ⚡ Ultra‑efficient runtime — smooth detection at 30+ FPS on mid-tier hardware.
You can also download the demo app showcasing real-time detection on device.

​
Integration Guide

​
1. Add the Maven Repository

Add the Dataspike SDK Maven repository in your settings.gradle.kts: 
dependencyResolutionManagement {
    repositories {
        google()
        mavenCentral()
        maven { url = uri("https://raw.githubusercontent.com/dataspike-io/Deepfake-MobileSDK-Android/release") }
    }
}

​
2. Add the SDK Dependency

In your app’s build.gradle.kts, include: 
dependencies {
    implementation("io.dataspike:deepfake-sdk:VERSION")
}
Replace VERSION with the latest SDK release number.

​
Initialization

Before analysis, initialize the SDK with your valid license key. 
val deepfakeSDK = DeepfakeSDK.getInstance()

val success = deepfakeSDK.initialize(
    applicationContext = applicationContext,
    license = "YOUR_LICENSE_KEY",
    nativeLogLevel = 4,        // Optional: Log level (0–5)
    deepfakeThreshold = 0.5f,  // Optional: Adjust detection sensitivity
)
If initialization succeeds, the SDK is ready to analyze both video and audio in real time.

​
Video or WebRTC Analysis

You can analyze frames directly from a standard Android View or a WebRTC VideoTrack.
Provide an AnalysisCallback to receive frame-by-frame results. 
val analysisCallback = object : AnalysisCallback {
    override fun onFrameAnalysisResult(result: AnalysisResult, message: String?) {
        // Handle frame result here (e.g., show confidence score)
    }

    override fun onAudioAnalysisResult(result: String) {
        // Handle audio analysis result (if applicable)
    }

    override fun onAnalysisError(error: String) {
        // Handle any errors
    }
}

// Analyze a view
deepfakeSDK.startAnalysis(yourView, analysisCallback)

// Or analyze a WebRTC track
deepfakeSDK.startAnalysis(yourVideoTrack, analysisCallback)
Tip: For real-time streams, prefer WebRTC integration. For local files, attach a SurfaceView or TextureView with playback controls.

​
Audio-Only Analysis

To detect manipulation in live or recorded speech, you can start an audio-only session: 
deepfakeSDK.startAudioAnalysis(analysisCallback)
Feed raw PCM audio data (16 kHz, ShortArray) to the SDK in small chunks: 
deepfakeSDK.processAudioChunk(yourAudioData)
The SDK processes ~3 s (48,000 samples) per chunk. Shorter segments should be zero‑padded.

​
Stop and Clean Up

When analysis is complete or the user session ends: 
deepfakeSDK.stopAnalysis()
deepfakeSDK.shutdownWithTimeout()
This releases all SDK resources and gracefully closes background threads.

​
Performance Notes

  • Runs entirely on-device — ideal for offline or privacy‑sensitive apps.
  • The deepfake model is quantized and optimized for mobile GPUs, achieving real-time performance even on mid-tier Snapdragon and Dimensity devices.
  • Supports continuous analysis of video + audio streams with minimal impact on battery or CPU.
  • All computations are ephemeral — no frame storage or network transfer.
Typical latency: 10 ms per frame (mid-tier device, 720p stream).

​
Licensing

You must provide a valid license key during initialization.
Licenses can be requested through your Dataspike dashboard or account manager. If the SDK is not initialized or an invalid key is used, detection methods will fail gracefully with an error callback.

​
Demo Application

A fully functional demo APK is available for download:
👉 deepfake-demo-app‑1.0.0.apk This demo demonstrates:
  • Real-time video and audio deepfake detection
  • Frame-level confidence scores
  • Offline inference and responsive UI updates

​
Summary

The Dataspike Deepfake SDK enables developers to embed AI-manipulation detection directly inside Android apps — without sending data to the cloud.
It’s optimized for speed, accuracy, and privacy, providing true on-device protection against synthetic media threats.
🔒 Offline. Real-time. Private.
Detect deepfakes anywhere — no connection required.