Fingerprinting Smartphones Via Microphones and Speakers

Project description

The objective of this project was to determine if smartphones can be fingerprinted using the microphones and speaker embedded in them. The main idea behind this work lies on the fact that chips like microphones and speakers even if they are streamlined, some manufacturing level idiosyncrasies are bound to exist which can be extracted from low level signal features. We extract such features to uniquely distinguish 50 android phones. The details on the work can be found in our paper. Also we have technical report which includes some extended description.

Source code

The source code for our android app is available at app.

The code for feature extraction and selection, along with the code for classification is available upon personal request. However, we mainly utilize the MIRToolBox to extract all the signal processing features.

Brief summary of findings

We found that both microphones and speakers embedded in smartphones can be uniquely characterized from the collected and generated audio signal, respectively. Combining features from both microphones and speakers provide the best outcome in terms of uniquely fingerprinting the source device. Mel-frequency cepstral coefficients (MFCCs) were found to be the most dominant features in fingerprinting acoustic components.