Voice biometrics #

Identification and voice recognition are issues that have a number of practical applications in automation, authentication and security. It is a popular method of remote authorization thanks to its non-invasive and accessibility (e.g. telephone, personal computer). This method combines physical and behavioral elements. It uses a person’s acoustic features, which are shaped by biological features (e.g. the shape of the larynx) and by additional behavioral features such as stress, rhythm, intonation and vocabulary selection.

Pre-read (Required) #

• https://azure.microsoft.com/en-us/services/cognitive-services/speaker-recognition/#features
• https://docs.microsoft.com/en-us/azure/cognitive-services/speech-service/speaker-recognition-overview#speaker-verification
• https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.spectrogram.html

Class content #

• Implementation of audio sampling and visualization (e.g. MFFC, spectrogram).
• Introduction to 1D convolution and Recurrent Neural Networks.
• Implementation of Microsoft Azure - Speaker Recognition.
• Solution security analysis, attack vector through voice imitation, voice generation (deepfake).