Eric Tappert wrote:
>+---------------
>| BTW, voice detectors have an "attack time", so often the first part
>| of a syllable is clipped, reducing the quality of the connection. >+---------------
>Though note that digital voice detectors are capable of storing a
>FIFO's worth of samples from *before* the "on" decision is made and
>can include those in the outgoing stream ... at the cost of delaying
>the *entire* "on" period by that much [that is, by the attack time of
>the detector]. While increasing end-to-end delay is never desirable,
>increasing it by just enough to restore the clipped portion is
>probably worth it.
And a really, really, "smart" system of this sort would buffer that first 1/4 or so second, feed it out (as you described) as part of the delayed stream, and then...
... and then, "speed up" the next few seconds of output until it catches up with the speaker's "real time". [a] [a] There are numerous ways of taking a voice stream and reducing the time length. This is done by, for example, radio stations to cut a 25 minute (say...) interview or talk fest down to 20 minutes, thus letting them pump in more commercials.
For better or worse, Rush Limbaugh was the most vocal, so to speak, opponent of this and he got the stations to stop playing with his broadcasts.