^SynOhrMultiSenseKNX-Eng-13.odt, 2014-05-06 Page 45 of 49
Range and Master
Slave
The maximum range depends mainly on the room acoustics, the place of installation (e.g.
acoustic shadow behind a wall), the speaker volume and also (but less strongly) on the number
of used words. Up to a range of 4 m a everyday use parameterization should be possible even
under difficult circumstances. For the range it is important that it is frontally spoken or from one
side. At lateral coupling the signal over 50 % is worse – therefore a human being is turning
automatically his head towards the speaker. Additionally the similarity of the selected words as
well as the room acoustics are playing a major role. If the noise sources (radio etc.) are louder
and closer to the SynOhr as the speaker a recognition is impossible. If larger areas are
acoustically to be „listened“ it takes several slaves.
Room Acoustics
Relationships with resonances or room echo are often existing in hallways or tiled rooms and
without wallpaper. With unfavorable parameterization a recognition is possible over long
distances (7..10 m), but this is in practice not robust against inference noises. The parameter
„Room Size“ in SynOhrStudio (comp. page Fehler: Referenz nicht gefunden) adjusts the range.
Finding the optimal range under difficult conditions do not change any other parameters, but
choose the range small in order to adjust gradually the desired sensitivity. At high sound level
the tendency of distorting the signal is internally limited by a limiter. This allows a stable
operation for seldom overrides. If the limiter is constantly active so the keyword recognition
responds on wrong words or is generally not robust. Therefore please avoid a „too large“ room
size.
Commands
The recognition of commands is processed differently than that of the keyword. Here the
recognizer chooses the solution that comes closest to the configured command, this is a
similarity strategy. The strategy provides a great tolerance of speech of commands at their
recognition. If very few commands are used, the tolerance is automatically greater. For example
„LAMELLA DOWN“ can be initiate by the statement „A-HA“. Because of the similarity strategy it
is important that the keyword recognition is robust.
A model representation of the similarity strategy is that of a human being who learns a new
language, but is just knowing „LAMELLA DOWN“. He can only differentiates slightly similar
words more severe than a human being who knows the total vocabulary of a language.
To parameterize the similarity strategy, please change in SynOhrStudio the experts setting
„Speech Quality Command“. If the SynOhr MultiSense KNX should work only with one or two
commands and the similarity strategy is still bothering although the corresponding
parameterization, then we recommend to fill the „unused“ commands with unused words.
The higher the echo in a room is the more important is a good quality of recognition, that the
speaker inserts between each word of a command sequence a little break (<0,4s). This is also
dependent on the similarity of the used words respectively increases a break at similar phrases
distinctness of the words. The recognition quality is to be understood in context of numbers with
their similarity such as between FORTY and SEVENTY in German. The user can improve this
problem by significant pronunciation of the single words even at greater distances.
Enertex
®
Bayern GmbH – Ebermannstädter Straße 8 - 91301 Forchheim - [email protected]
