New Model !! Better Accuracy !!

As per many of our community members' requests a new model named Resnet_50_Arcloss is trained and some api changes too are incorporated to facilitate the same Moreover the newer model's input length is increased to 1.5 seconds to facilitate longer words

Kindly try it out and let us know

Lets GO!!! : )

The engine used a sliding window approach to listen for hotwords, this results in one utterance having multiple triggers. To minimize the same, some complicated logic was used resulting in unnecessarily complex api. In this release we shift to more simpler approach i.e relaxation_time (min time required between any 2 triggers , earlier triggers will be dismissed) , resulting in more simpler programmer api

However these large upates are breaking changes : (

Improved false positive reduction and changes to reduce multiple trigger per utterance of the hotwiord to one

TODO: Need to update documentation accordingly