Microsoft shows off its weird Silent Voice technology

Microsoft is working on a new voice input interface that allows users to speak and record without the presence of sound. The research was conducted by Microsoft Research and presented at ACM CHI 2018. The technology, called SilentVoice, enables communication by recording the sounds made while breathing, which allows whisper-like sounds to be enough for the microphone to record speech without disturbing people around. In addition, the module will also filter out surrounding speech, so users can capture clear speech even with external interference.

SilentVoice is a new voice input interface device that enables voice-based Natural User Interfaces (NUI) in everyday life.

The proposed "progressive speech" method enables the microphone to be placed very close to the front of the mouth without being affected by noise, capturing very soft speech with a good S/N ratio. It achieves ultra-small (less than 39dB(A)) speech leakage, allowing the use of voice input without annoying people around in public and mobile situations, as well as in offices and homes. (Finally, it won't bother people using TNT!)

By measuring the direction of airflow, SilentVoice can easily separate external sounds from normal speech with an accuracy of 98.8%, and no activation words are required before voice communication starts. It can also be used with a voice activation system with a specially trained speech recognizer. The evaluation results produced word error rates (WERs) of 1.8% (speaker-dependent conditions) and 7.0% (speaker-independent conditions), including 85 command sentences, which means that natural speech similar to whispers can also be used for real-time voice communication.

You can view the full presentation at the ACM CHI Conference on Computing Systems: https://youtu.be/9EV1mEtVfuM

The technology is still in the research stage but will definitely help those who like to use voice commands but prefer to work without disturbing those around them.