UPDATE July 31, 2013: Supertoy Kickstarter reaches £30,000 goal, with 22 days to spare, Ashley Conlan announces. I can’t wait to talk to (and hug) Teddy when he arrives. “I don’t know what he is going to say for sure.” (It does have a blacklist for NSFW words though.) “At the end of a demo, when I said ‘Bye Now,’ Supertoy responded, ‘Will that accomplish your objective?’ That was totally a surprise.” However, “Supertoy is not scripted,” said Conlan. For example: a child says, “Teddy, let’s play cowboys.” Teddy role-plays like a cowboy, says things like “howdy partner,” and when the child has finished playing role play, he or she says, “Teddy, stop playing cowboy.” Teddy outfits (cowboy, wizard, marine, footballer, fisherman, etc.) will also be available. ![]() ![]() Supertoy Robotics also just announced Sunday a new role-playing feature. It reads bedtime stories, sings songs, and answers all those pesky ‘why’ questions.” Being connected to the Internet also means Supertoy upgrades over time and becomes better and better as it learns. “This allows children to learn from the vast amounts of educational material online, without parents having to worry. “It is designed to be used in a safe and child-friendly manner, so only age-appropriate information is shared. “Supertoy is fun and playful for children and can also be their first introduction to the world of the Internet,” said Conlan. You can also ask it to make calls for you, give you a wakeup-call, find the nearest restaurant, launch apps, and other Siri-like stuff, and it “safely searches the web,” said Conlan. Like Jeannie, you can ask questions or engage in conversation, and actually learn things. ![]() (I’ve been “talking” to Jeannie for a few days on my iPad - and on an Android Nexus 7, where she’s more developed - and I’m impressed.) You can get a rough idea of how Teddy works in this impressive demo of Jeannie (formerly Voice Actions), but Teddy is even more realistic. Teddy “evolved” from Jeannie, a popular Siri-like app for iOS (iPhone, etc.) and Android, said Karsten Flügge, with more than 3 million downloads, created by Pannous (of which he is CEO). Your question or comment then goes to the Supertoy Robotics server, which generates a natural-language audio reply that goes (via the smart phone or device) back to Teddy’s speaker. Assuming you have a smartphone data plan or WiFi, the server detects your language and converts your speech to text. “The magic is in the software, not the robot,” Conlan explained to me. Teddy connects via standard 4-segment audio plug to an iPhone 4+, iPad2+, iPod touch 3G+, or Android (2.3 +) smartphone or device - Conlan suggests putting the phone inside the Supertoy for realism. The robot’s hardware is simple: just an audio in/out interface and a motor for mouth animation. It has “evolved” from Supertoy Robotics’ popular Jeannie chatbot (for iPhone, iPad, and Android) - but is far more advanced. The Supertoy technology is Siri-like, but better, because it’s not just a Q&A but a continuous conversation and Teddy will also remember past discussions. Once the graph is iniatiated using tflearn.init_graph(), it can then be defined.Supertoy Robotics’ Ashley Conlan, Kartsen Fluegge, and friendĪccording to England-based Supertoy Robotics’ CEO Ashley Conlan, Supertoy can speak 30 languages out of the box, and because it gets its smarts in real time from the web, the company can easily upgrade it and do updates, he explained to me. Voice Actions, Pannous, chatbot, virtual assistant, chatterbot, virtual agent, chat bot, conversational agent, brand agent, virtual employee, 9897. ![]() The actual deep learning is done using TFLearn. These are put into the variables X and Y, respectively. The output of the function is a collection of wav file batches and their associated labels. Android: Jeannie, formerly Pannous Voice Actions for Android (not to be confused with Google Voice Actions) has always been a solid voice control utility for your Android phone, but the developers. The source variable is used to download the data if it is not already present in the folder. Next, the function shuffles the dataset and then creates different chunks of wav files that are used to fill the batches until the size is equal to batch_size. NumberSpoken_PersonSpeaking_SampleRate.wavĮssentially, assuming the data is already downloaded, the script does the following:Ĭreates an array of the different speakers in speakers, where the length of this array is the number of classes used for training.ĭivides the training data set into different batches using batch=data.wave_batch_generator(batch_size=1000,source=WORD_WAVs,target=)įirst, the Target of the script tells the function what the target of the parent script is: either speaker classification or digit classification.
0 Comments
Leave a Reply. |