[Elecraft] Future KX3/KX2 accessories?

[David Woolley]

Computerised language translation is predicated on good continuous 
speech recognition.  For a long time that is going to rely on relatively 
clean audio.  Even if automated translation in noise becomes possible, 
with the amount of noise, etc., on an amateur radio circuit, the 
additional degradation in going from digital to analogue and back is 
likely to be negligible.

There is also a fundamental problem with machine translation in that the 
solution for the translation will not stabilise until the end of 
sentence, or later.  Whilst a human also needs to reach this point for 
full understanding, they will have part processed the speech before they 
get there.  In the general case, the machine translator cannot even 
start passing the translation on to the human until it reaches this 
point, so there will always be a significant extra processing delay, 
compared with understanding the language, directly.

I would also expect any software or hardware that comes onto the market 
to have been designed for use with at least the telephone bandwidth of 
300-3.4kHz, not the narrower bandwidth used for SSB radio.  Even 
telephone bandwidth is not enough to accurately recognize sibilants (s, 
sh, h, etc.).

The main case where direct digital is useful is for digital mode, where 
phase errors, which have no impact on speech recognition, may be 
significant.

On 30/03/17 19:20, Steve Sergeant wrote:

> How abut signal processing operations that
> are beyond the DSP capability in the radio? How about some
> not-so-distant future when spoken language translation might be possible?