independent 2. Sta)s)cal Model Of Language (SML): free language-‐ speaker independent. • Needs InformaHon ExtracHon(IE) NLP to preprocess the ASR output to get semanHc informaHon.
$juego; $juego = ( "adivinar personajes":AKINATOR | "habla mas bajo":BAJARVOLUMEN | "callate ya":CALLARSE | "no sigas diciendo mas cosas":CALLARSE | "escondite":ESCONDITE | "quien soy":ESCONDITE | "ahorcado":HANGMAN | "deja de hacer esto":PARAR | "dejalo ya":PARAR | "estate quieta":PARAR | "callate un momento":PAUSESPEECH | "reconocer animales":PELUCHES | "reconocer peluches":PELUCHES | "peluches":PELUCHES | "jugar con los peluches":PELUCHES | "animales":PELUCHES | "jugar con los animales":PELUCHES | "ya puedes seguir hablando":RESUMESPEECH | "persigueme":SIGUEME | "pilla pilla":SIGUEME | "seguir":SIGUEME | "sigueme":SIGUEME | "vamonos de paseo":SIGUEME | "vamos a dar una vuelta":SIGUEME | "vente conmigo":SIGUEME | "habla mas alto":SUBIRVOLUMEN | "tres en raya":TICTACTOE | "habla lo mas alto posible":VOLUMENMAX | "habla lo mas bajo posible":VOLUMENMIN) {<@Game $value>}; Example of Seman)c Grammar
Natural language (built-‐into word out of typical dicHonaries, but present in “street vocabulary”). – Needs a postprocess phase. This try to get the semanHc informaHon of the text transcribed using Natural Language Understanding => InformaHon ExtracHon, AbstracHng, etc.
Cancela)on => Necessary dedicated microphones and powerful algorithms. – Ac)ve Echo Cancela)on: remove the robot or computer sound of the signal received. – Voice Ac)vity Detec)on (VAD): determinate voice and unvoice audio samples (when the audio-‐voice start and end). ASR built-‐into very simple VAD system (designed for telephonic applicaHons) based on volume threshold. • ¿Push talk or VAD? – Sound Source Separa)on (SSS): split the audio signal on several channels. Each channel corresponds to a peer.