The whole experience has made him more aware of the digital generational gap while Felisa finds it strange to press a button to record and send an audio message, younger generations take these procedures in their stride.īut everything has a cost-effective solution: Lucio is thinking of replacing the button that connects to the microphone with a traditional phone, because Felisa is in the habit of picking up a handset and talking into the mouthpiece. We had a proper, flowing conversation that was impossible before Manuel Lucio, Spanish engineer The computing aspect of the Grannygram is executed through a program that Lucio has created himself combined with the existence of programming languages in Telegram, collected in a virtual library on the Python platform. Lucio says he likes to use local shops for components, thereby ensuring the quality is decent and also completing the cycle of local entrepreneurship for those who enjoy “tinkering” in their spare time. The printer cost around €30, the cables and connectors, €10, the microphone was €6 and the Raspberry Pi – a micro computer that organizes the whole procedure – set him back another €30. Lucio’s solution might have come at a price regarding brainpower, but in terms of euros, it was cheap. While Felisa’s arthritis prevents her from using screens and telephones, hearing problems make it uncomfortable for her to chat on a cellphone. A printed message comes out on the Yayagram. “We had a proper, flowing conversation that was impossible before,” he says. The engineer laughs when he recalls the first test he did with his grandmother, which involved telling her they were having artichokes and ham for dinner. It also has handles so that its owner can carry it around if she needs to. The machine uses a WiFi internet connection and is plugged into an electrical outlet just like any household appliance. Lucio explains that his grandmother doesn’t quite understand how the contraption works but is “delighted” all the same. When Felisa receives their reply, she can read it in enlarged print thanks to a small thermal printer “like the ones in supermarkets,” which is connected to the Yayagram. The app tells them, “Yayagram has a message for you!” and they can then listen to their grandmother. Her message is then sent as a normal audio, and reaches the Telegram messaging app that her family members have downloaded. Lucio, 35, defines his device as a “contraption” and avoids using any fancy words to talk about “a box that allows communication via voice messages.” The box has a button that Felisa presses to start recording her voice, and then, just as telephone operators used to do, she selects who she wants to contact. A Telegram message created by the Yayagram Now, thanks to his ‘Yayagram’ – which plays on the Spanish informal term for grandma, yaya – the family matriarch can send and receive messages all by herself. So Lucio put on his thinking cap and came up with a solution that has earned him thousands of Twitter followers. Her family is scattered between Burgos, Oviedo, Bilbao and Madrid, and many relatives cannot go see her in person because of coronavirus-related restrictions on movement. And then there is Manuel Lucio, a Spanish computer engineer who came up with a nifty invention because his grandma had trouble using the phone.Ī resident of Burgos who suffers from arthritis, Felisa Romano, 96, was finding it increasingly difficult to talk to her grandchildren on the phone or via video conference. 6th International Conference on Spoken Language Processing (ICSLP 2000), vol.Some great entrepreneurs got their start in a garage. (2000) Cross-domain robust acoustic training. The acoustic model is evaluated on both flat grammars, digit and name at department, and language model tasks, ATIS and general dictation, using the IBM ViaVoice product engine.Ĭite as: Jan, E.-E., Botella Ordinas, J. Based on these studies, we are able to develop a hybrid system for 8KHz closing talking microphone, telephony landline and cellular phone environments. We then investigated the performance degradation on two reduced phone sets for Spanish. The results show that by fixing number of parameters, system with smaller number of context dependentHMMstates yields better accuracy. We explored the effects on decision tree size to accuracy by approximately 10%. We used weighted multi-style training by pooling insufficient telephony landline and cellular data with down sampled wide band clean data to develop better hybrid acoustic models. This paper describes our efforts towards cross-domain acoustic training for LargeVocabulary Continuous Speech Recognition (LVCSR) systems. Cross-domain robust acoustic training Ea-Ee Jan, Jaime Botella Ordinas