[nicolasehrhardt]

I don't think a small team of people can be better than Google or Apple in building a speech to text technology. However, leveraging the tools available to help the deaf is the main idea - at least at the beginning - I think. Going forward, they will probably focus on their "Leap Motion" part of the project: from signs to text/voice and let big companies improve their text to speech algorithms that they would just use. Because that's where big improvement can be made.

[cerrelio]

This technology (speaker identification) is 10 years old, and [HMM/neural net] speech recognition is slightly older. So a small team could likely pull it off today just by implementing or using code published by researchers. As long as Transcense have control over the microphone(s), then it might work. Single mic/multispeaker speech recognition is still practically impossible unless the speakers take turns (not always the case).

[abecedarius]

The Leap Motion thing is unrelated, from some other group, at least according to the story.

The problem of supporting realtime conversation among multiple people is different enough from voice search that there's scope to differentiate.