Hide and Speak: Deep Neural Networks for Speech Steganography

Felix Kreuk*, Yossi Adi*, Bhiksha Raj, Rita Singh, Joseph Keshet

*Corresponding authors.

This post presents "Hide and Speak: Deep Neural Networks for Speech Steganography", a deep steganography model for speech.

Steganography ("steganos" - concealed or covered plus "graphein" - writing) is the science of concealing messages inside other messages. It is generally used to convey concealed "secret" messages to recipients who are aware of its presence, while keeping even their existence hidden from other unaware parties who only see the "public" or "carrier" message.

In this paper we propose the use of deep neural networks as learnable steganographic functions, that learn to optimally exploit redundancies in audio data to conceal messages. We demonstrate quantitatively that the proposed method is able to both effectively hide secret messages into a carrier and recover them from the carrier. Qualitative experiments suggest that modifications to the carrier are unnoticeable by human listeners and that the decoded messages are highly intelligible as well.

Additionaly, this scheme now permits us to hide multiple secret messages into a single carrier, each potentially with a different intended recipient who is the only person who can recover it


Italian Trulli

Here are some samples from our steganography system for you to listen. "Original Carrier" refers to the original audio in which we conceal a hidden message. "Embedded Carrier" refers to the audio with a hidden message, "Original Message" referes to the original message we wish to hide inside "Original Carrer" and "Decoded Message" is the message decoded from "Embedded Carrier".

Single Message

Original Carrier Embedded Carrier Original Message Decoded Message

Three Messages

Original Carrier Embedded Carrier Original Message Decoded Message








Five Messages

Original Carrier Embedded Carrier Original Message Decoded Message
















Code will be available soon!