Call3 Monitoring SPEAKY

Public Summary Month 4/2013

Tue, 30 Apr 2013 22:26:32 GMT

In the last part of the experiment, our activities have focused on the testing and validation of the prototypes implemented and on dissemination activities.

UNIRM1 has developed an innovative Web-based corpus collection infrastructure, in order to attract a very large community. This initiative has been announced to the RoboCup@Home community, and the output of this collection will be an extensive corpus for human-robot interaction within home environments, that will be published for the whole robotic community and for RoboCup competitions.

A second major activity of this month has been a set of usability study conducted within a real home environment, to evaluate the whole speech system with twelve English native tongue speakers. The results of the experiment are currently being analysed.

In addition UNIRM1 and MV have cooperated to the evaluation of the wizard developed by MV with the second scheduled demonstrator: surveillance.

Finally, demonstrations of the Speaky 4 Robots experiment have been presented at “Robots on Tour” (Zurich, March 9^th) and at “RomeCup 2013” (March 23^rd, Rome) by UNIRM1. UNIRM1 and MV are both committed in disseminating the outcomes of the Speaky for Robots project in the next months.

Public Summary Month 2/2013

Tue, 26 Feb 2013 23:52:50 GMT

February has been completely dedicated to the evaluation of our system with external non-expert users (Task 4.3), and to additional testing of the wizard component (Task 4.2).

UNIRM1 has worked on the design of an experimental evaluation with non-expert participants. Based on the experience acquired through the first exploratory evaluation.The whole process has been split into a new lexical corpus collection, this time based on a web-based architecture to guarantee access to a large audience, and the setup of a usability study involving 12 people that will take place in April.

MV has successfully evaluated the new wizard component through internal validation and testing.

UNIRM1 will present Speaky for Robots at “Robots on Tour”, on March 9th 2013, and at RomeCup, on March 22nd 2013.

Public Summary Month 12/2012

Fri, 28 Dec 2012 10:20:45 GMT

In the last bimester, we have been working on the experimental evaluation of the prototypes (Task 4). On the one hand, we have defined and tested the procedure for performance evaluation of the vocal interfaces defined for different robots and environments. On the other hand, laboratory experiments have been performed to improve the prototypes.

MV has finalized the development of the wizard prototype (Task 3.2) and the corresponding technical report.

S4R has been presented at Robots @ DIAG November 26^th, at DIS (Roma) within the events organized for the EU Robotics Week 2013.

Public Summary Month 10/2012

Tue, 30 Oct 2012 13:30:55 GMT

In the last bimester, we have focussed our work on with the refinement and testing of the Service Prototype and Demonstrator, (Deliverable 3.1) which is part of the Task 3: System Development, Implementation, Integration and Testing. Specifically, we have completed the interface to the MS dictation engine, in preparation to a refinement of the ASR process, that relies on both a recognition based on a language model specified using an application dependent grammar and a generic language model. We have adopted and specialized the dialog infrastructure offered by the Speaky Mediavoice development environment and we have thus delivered a new version of the prototype system. We have run some tests on the VUI, and we have collected the results obtained by processing a small corpus of spoken sentences. The analysis of these results has been used in order to define possible improvements of the prototype.

S4R has been presented at the ECHORD Workshop in connection with IEEE IROS 2012, Vilamoura, Portugal, October 11^th, 2012.

Public Summary Month 8/2012

Tue, 28 Aug 2012 15:59:27 GMT

In the last bimester we have worked on with the activities of Task 3: System Development, Implementation, Integration and Testing, namely on the implementation and testing of the first prototype of the VUI for the chosen domains. This effort will be reported in Deliverable 3.1, Mid-term Home Service Prototype and Demonstrator, Mid-term Home Service Prototype and Demonstrator. More specifically, we have built a debugging interface for the speech interpretation process. Moreover, we have implemented in ROS the modules for the integration of the knowledge about the environment, represented as a semantic map, and for “grounding” of the commands of the robot in a real application scenario. Finally, we completed the preliminary work on Task 4: Experimental Evaluation; on the definition of experimental set-up and performance metrics, by acquiring the semantic map of a real home, that will be used to display the execution of the commands given to the robot using the Player/Stage simulator. The poster describing the current state of the project has been successfully presented at AAAI-12 Media Fair (Toronto, Canada, July 23-27).

Public Summary Month 6/2012

Thu, 28 Jun 2012 15:08:13 GMT

In the last bimester we have worked on with the activities of Task 3: System Development, Implementation, Integration and Testing, namely on the implementation and testing of the first prototype of the VUI for the chosen domains. According to the results of the design phase, we are focusing on the generalized components that are the basis for the implementation of the VUI and development of the wizard is moved to a later stage. More specifically, we have a complete modular grammar for the humanoid robot in the home domain, according to the methodology described in the appendix of the Technical Report D21. To this end, we have updated the command description therein, since we have now released a more polished and complete set reference for the commands of the robot. In addition, we have refined the software architecture and implemented a simple prototype version of the components for selecting the interpretation and managing the dialog with the user. Moreover, we have started the work on Task 4: Experimental Evaluation; in particular, we have begun working on Task T4.1: Definition of experimental set-up and performance metrics. Finally, a poster describing the current state of the project has been accepted for presentation at AAAI-12 Media Fair.

Public Summary Month 4/2012

Mon, 30 Apr 2012 20:36:13 GMT

In the last bimester, we have consolidated the S4R activities related with Task 1, and we have completed the S4R activities planned in Task 2, concerning the design of the process for the generation Vocal User Interface for specific robotic applications. The details are available in the Technical report released at the end of Task 2 (April 30^th). In March, S4R decided to discontinue the development with the ASR Loquendo (due to unavailability of the product), and adopted the Microsoft Speech Platform, which required an additional integration effort, after which the new platform became functionally compatible with the previous one. In addition, we have implemented the interface for testing our interface on the Cobot platform developed at Carnegie Mellon University, as an additional robotic platform to support a solution that can be easily customized to different robotic platforms. In March, the prototype implementation of S4R on the NAO platform has been demonstrated at RomeCup 2012 (a RoboCup Junior event, held on March 23, 2012 in the Rome City Hall) and tested by the vice mayor of Rome.

Public Summary Month 2/2012

Wed, 29 Feb 2012 18:37:57 GMT

We have built two example collection of commands to model the two application domains. The sentences have been collected from the students of the class “Seminars AI and Robotics”, section on Natural Language Processing. The goal was experimenting the process of acquiring domain specific information, and the collections represent a first characterization of domain specific knowledge, in terms of robot capabilities, environments and tasks, not the final resources for language generation. We have combined the collection of the lexical units defined in the RoboFrameNet environment (available in Ros), with the more general representation provided by Wordnet and defined a first characterization of the possible wordings of the commands.

We have defined a methodology for testing the performance of our system. The baseline will be a system where a set of sentences from a corpus is associated with a set of commands. The performance of the customized system will be compared through systematic experiments with the baseline, on a set of vocal data recorded from different users. A dedicated tool has been built in order to support the systematic evaluation of the system performance.

The first technical report provides a comprehensive report of the first activities carried out in the project.

Public Summary Month 12/2011

Thu, 29 Dec 2011 15:20:24 GMT

Experiment started in November still pending the signature of the contract.

Activities have focussed on the analysis and initial steps in the design. Speciifcally: Robotic Domain Definition and Representation and Robotic Voice Development Kit Design

The main steps accomplished in task 1 are the implementation of a prototype to interact using Speaky both with the NAO humanoid robot by Aldebaran and with a small wheeled Erratic robot by Videre Design. The goal is to acquire some practical experience on the possible wordings used to command the robot. In the implementation of the above mentioned prototypes we have provided a first characterization of the robot capabilities, both for a humanoid robot and for a wheeled robot.

The main steps accomplished in task 2 consist of a first architectural design in the implementation of the above described prototype, which is specifically concerned with the communication between Speaky and the mobile platform.We have also identified the internal structure of the speaky devolpment environment and identified two key components: the dynamic grammar manager and the the dialogue manager. The first one is instrumental to provide suitable data for a linguistic post-processing of the output of the speech recognition and the second one needs to integrate the state of the dialogue with the state of the robot.