Group items tagged

Free a intégré l'assistant vocal Alexa à sa nouvelle Freebox Delta. Il s'agit donc, aussi, d'une enceinte connectée et intelligente. Un bouton est présent pour activer/désactiver l'assistant vocal. Free a aussi inventé : "Ok Freebox". On peut ainsi contrôler son téléviseur à la voix.

Passionnant décryptage du CTO de Snips... depuis Siri à aujourd'hui... décryptage de la technologie autour des assistants vocaux, cas d'usages, fantasmes... Interface vocale, la technologie existe depuis 30 ans (médecins, call center...), "interface" remplacée par "assistant" (promesse de l'IA) + promesse d'un dialogue naturel... Limites : la voix ne sert pas à tout, besoin d'un écran quand on a une recherche plus précise à faire... ex: est-ce à l'usager de s'adapter à l'interface ? = > formulation de l'intention... Briques techniques d'un assistant vocal : - banc de micros (détection de la source, triangulation pour mieux capter la source de son) - wakeword pour réveiller l'assistant (machine learning = écoute le signal en permanence mais entre en écoute active à partir du "wakeword", la reconnaissance vocale est alors lancée qui a pour objectif de transcrire ce que l'utilisateur a dit) - NLU : natural language understanding https://en.wikipedia.org/wiki/Natural-language_understanding prend le texte en entrée ("la météo demain à Bordeaux ?") et sort un objet structuré : intention : météo, quand : demain soir, où : bordeaux même fonctionnement qu'un chatbot - Logique d'action (comment réagir à la demande de l'utilisateur ?) :ce sont les développeurs qui travaillent cette dernière partie + synthèse vocale pour donner la réponse Snips travaille les étapes "wakeword" + "NLU" logique d'action (cf. modèle app store) avec 16000 développeurs + synthèse vocale : sous-traitance secteurs : bureau, habitat, véhicule, industrie Privacy / snips : le son ne sort pas du "salon" - sortie prévue fin 2019 petit réseau de neurones pour le "wakeword" tourne en embarqué pour toutes les solutions (amazon, google...)... mais il y a un taux d'erreur (de une fois par heure à une fois par jour)... donc

> la BBC ne met plus ses podcasts à disposition des services google podcast ou google assistant "You might have also seen that our podcasts are no longer available on certain Google products - including the Google Podcast app and Google assistant. I want to explain a little bit about why that has happened. Last year, Google launched its own podcast app for Android users - they've also said they will launch a browser version for computers soon. Google has since begun to direct people who search for a BBC podcast into its own podcast service, rather than BBC Sounds or other third party services, which reduces people's choice - an approach that the BBC is not comfortable with and has consistently expressed strong concerns about. We asked them to exclude the BBC from this specific feature but they have refused. As a public service, we want our content and services to be available to as many people as possible and we make these available for free on a range of third-party apps. But as the BBC, funded by the licence fee payers in the UK, we have to ensure it is done in a way that is good for all audiences, according to our Distribution Policy - which has been agreed with Ofcom." "We also want to make our programmes and services as good as they can possibly be - this means us getting hold of meaningful audience data. This helps us do a number of things; make more types of programmes we know people like, make our services even more personalised and relevant to people using them, and equally importantly, identify gaps in our commissioning to ensure we're making something for all audiences. Unfortunately, given the way the Google podcast service operates, we can't do any of the above."

"Los Angeles medical center Cedars-Sinai is currently piloting a program that places Amazon Echos in more than 100 patient rooms. The smart speakers use Aiva, a voice assistant platform for healthcare, and is intended to help patients communicate with their caregivers. Letting patients use Alexa to perform basic tasks like changing TV channels also frees up nurses to perform medical care."

podcast CES France Info sur les assistants vocaux français Linto, Vivoka et Snips questions sur la conception... pour des assistants vocaux "gafam free" Linto & Snips btb pour l'instant Vivoka : btbtc (hôtels)

"Spotify is about to begin publicly testing its first hardware: a voice-controlled smart assistant for cars, meant to help Spotify learn how people consume audio while they're driving. As part of the test, some Premium users will receive the device for free. The device, called Car Thing, plugs into a vehicle's 12-volt outlet (aka a cigarette lighter) for power and connects to both a person's car and phone over Bluetooth. The device can be activated by saying, "Hey, Spotify," followed by a request for whatever the person wants to hear. It'll be linked to a user's Spotify account, so they can access their playlists." [...] "While smart assistant companies want access to drivers' in-car data, carmakers themselves are developing their own smart assistants and voice controls, too. But in a survey by JD Power, 76 percent of car owners said they're interested in having the same brand voice assistant in their home and in their car. Spotify could face an uphill battle if it wants to convince its users to give up the assistant they're already used to in order to play audio more seamlessly from their Spotify account."

"A new version of Monopoly doesn't use paper money, instead it has voice banking which will keep track of transactions."

Interview Luc Julia - créateur de Siri - 10'43" "L'IA n'existe pas" > "le marketing a kidnappé le terme d'intelligence artificielle, pour raconter n'importe quoi" - promesses dès les 60's ex : langage naturel et ont échoué, le 1er hiver de l'IA lA générale > il n'y en aura pas avec les techniques elles-mêmes qui n'ont pas changé depuis les 60's Mais grâce aux data, progrès dans les méthodes stats, des outils, qui vont nous augmenter mais pas nous remplacer. "IA possible" = statistiques et mathématiques ex: planification / mieux observer / motifs dans l'adn... Il ne faut pas promettre la lune... Témoignage Siri > "stupidité artificielle pour imiter l'humain", "Siri peut se tromper" parce que ça ne marchait pas bien, on a créé le "1er Siri, "the assistant" un peu sourd et fatigué, pour humaniser cette IA..." "Film Her : vue de l'esprit" Adoption de Siri parce qu'elle a été humanisée... produit sympathique auquel on pouvait poser des questions... but au départ amener la techno aux gens..." "depuis 2016, google home et alexa, font des choses sympathiques, ont fait des progrès malgré les pb de privacy" "les techno utilisées ont bcp évolué par rapport à la version de Siri de 2011" "Utilité des AV ? négo entre ce qu'on donne et ce qu'on reçoit / privacy" Les assistants 'ubiquitous" peuvent aider plein de gens. Ses travaux chez Samsung : santé, transport, objets intelligents (ex IOT)

"In a free service, bots call restaurants and make reservations. The technology is impressive, except for when the caller is actually a person." "In other words, Duplex, which Google first showed off last year as a technological marvel using A.I., is still largely operated by humans. While A.I. services like Google's are meant to help us, their part-machine, part-human approach could contribute to a mounting problem: the struggle to decipher the real from the fake, from bogus reviews and online disinformation to bots posing as people." "Google's A.I. is eerily human, when it works" "Duplex needs lots of data to improve" "Duplex is proficient at making a restaurant reservation over the phone, but much like Facebook, Google still leans on human intelligence. At any given moment, it is lifelike. But it struggles to deal with the unexpected. "There are three things that are important when it comes to A.I.'s interactions with humans: context, context and context," said Jerry Kaplan, author of "Humans Need Not Apply: A Guide to Wealth and Work in the Age of Artificial Intelligence" and a Stanford University lecturer on artificial intelligence. "Machines are very good with detail but terrible at context," he said."