At Amazon’s big hardware event last September, the Voice Interoperability Initiative (VII) was introduced to create an industry standard. This AI breakthrough gets Amazon one step closer to achieving Jeff Bezos’ vision for a world where intelligent personas can interact with each other cross-platform. We also used multi-target learning to mitigate the need for training a model for each wake word separately and multi-locale training to enable the new multi-wake word model to accept a wide spectrum of nuances in pronunciations.”Īs for future applications, Sindhwani added, “The innovation in multi-wake word detection is significant because it’s the first step toward enabling multiple agent interactions and accessing different personas across devices.” To accomplish this, he explained, “We were able to develop transfer learning and data augmentation techniques to train a new multi-wake word model that works well for all supported wake words and conditions. With “Alexa” having billions of customer call dialogues to train on but “Hey Samuel” relatively few, a multi-wake word model needed to be created that could make predictions with little to no data. Manoj Sindhwani, VP of Alexa Speech shared with me the engineering challenges his team had in getting the two word phrase “Hey Samuel” to co-exist with single wake words Alexa, Amazon, Echo and Computer. “Alexa, ask Samuel how he got his big break.” “Alexa, ask Samuel what his favorite superhero movie is.” “Hey Samuel, what’s your favorite TV show?” “Hey Samuel, where is your favorite vacation spot?” “Hey Samuel, what’s your favorite ice cream?” “Hey Samuel, what’s your favorite video game?” “Hey Samuel, what’s your favorite drink?” “Hey Samuel, what’s your favorite movie line?” “Hey Samuel, what’s the secret to your success?” “Hey Samuel, how did you get the part of Nick Fury?” “Hey Samuel, what was it like making Star Wars?” “Hey Samuel, did you always want to be an actor?” “Hey Samuel, how much wood could a woodchuck chuck?” “Hey Samuel, who is your favorite Star Wars character?” “Hey Samuel, how many times was the f-word said in Pulp Fiction?” “Hey Samuel, what’s your favorite skill?” “Hey Samuel, what do you think of snakes?”
0 Comments
|
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. Archives
March 2023
Categories |