Return to site


· Technology,Speech Recognition,testing

Having spent a few hours in an anechoic chamber this week for some testing, I got a new appreciation for the prevalence of echo and reverberation in our environment. Echo-less environments are not normal. Silence, at least to the extreme of an anechoic chamber, is not golden.

The moment the doors swung closed on the chamber and as you stepped further inside, you got the sense that someone put cotton in your ears. Your foot steps make no echo, you can’t tell if you’re getting closer to or farther away from a wall. If I were blindfolded, I’d have no sense of the volume of the room.

The dead sound of a close field mic or a recording room is nothing like how voice sounds in our day to day environments. After time spent in the chamber, you definitely know now what the reverb that affects far field performance sounds like.

It’s likely that the capability of devices to work for far field voice is going to increase dramatically over the next year. The major drivers are:

- A plethora of new hardware audio processing devices
- Training of speech engines for far field models
- New software for far field that can run on the application layer

All Posts

Almost done…

We just sent you an email. Please click the link in the email to confirm your subscription!