Work package 2: Dissemination and Exploitation

This is a short overview of important activities that have taken place:

A presentation was given at the Residential Expert Group meeting in Berlin (2008-06-12) addressing 8 global telecommunication companies.

OFFIS had a meeting with a local ISP provider (EWETel), presenting the project itself and the “ServerSASP” business case. EWETel is very interested and further activities are currently discussed.

The demonstrator has been shown at several occasions. Among others, the HIC platform has been presented in the showroom of INTEL Germany, where the demonstrator will be shown at least for half a year. One further highlight was the demonstration at the euroITV conference, where the demonstrator was shown in addition to the scientific presentation. The demonstrator then has been selected to be shown in an ORF TV news broadcast.

The business cases were refined and redefined during a second business opportunities workshop held in Madrid at TID premises on 2008-10-27 and 2008-10-28. The project will now focus on the “HIC on RG” business case with a fallback on the initial “HIC on STB” business case: HaH is outstanding in its approach to combine the use of the TV for communication with the environment, e.g. watching TV or videos, phoning, interaction with the home with a set of complementing assistive technologies for the typical situations of the average hearing impaired user.

The website has had major reworks and is now based on a CMS which facilitates updating and makes it easier for visitors to access new information. All project publications will be made visible on the web site

Work package 5: Synface Enhancements and Adaptations

Activities in WP5 have focused on finalizing the SynFace software prototype for integration into the HIC-platform. This has involved development, debugging, and optimization to meet the run-time requirements of the platform. Experiments have been carried out with speed/performance tradeoffs for the German recognition network, evaluating different levels of ANN connection pruning. For the non-verbal information extraction, real-time pitch extraction algorithms have been implemented and tested.

Preparation for an extended user testing has been another important activity. This included discussions regarding final details of the measurement setup, as well as hardware and software preparation for user studies..

The comparison of broadband versus narrowband versions of the SynFace recognizer has been completed. Two versions of the SynFace recognizer were trained on material from the Swedish SpeeCon database – one with a sample rate of 16 kHz and one downsampled to 8 kHz. These networks were in turn compared to the original SynFace 8 kHz recognizer, trained on the Swedish SpeechDat telephony database. Performance was evaluated using framerate measurement over a common broadband dataset, as well as through an SRT-based audiovisual perception test (Hagerman et al., 1995) with five normal hearing subjects. Results show significant improvement for broadband over narrowband, and for narrowband over audio-only conditions, while the original SpeechDat was not significantly better than audio-only conditions for this particular sentence set.

The results have been presented at SCLT’08 in Stockholm in November. Other dissemination activities related to SynFace during the fall include a paper and poster presentation at Interspeech 2008 in Brisbane, Australia as well as presentation at national meeting arranged by the Swedish Institute of Assistive Technology.

Work package 6: Global SASP Strategies

Noise reduction algorithms for the TV signal have been researched and implemented. Most effort went into the research of a novel binaural noise reduction scheme, which suppresses parts of the audio signal that differ in both stereo channels. An audio-based classifier has been researched and developed. Several different audio signal features have been tested using training material to achieve best differentiation between pre-defined audio classes. A video-based commercial break detection scheme has been implemented in Matlab

Extensive user evaluations of noise reduction algorithms developed in WP6 have been prepared and started with four groups of subjects (about 60 in total) with mild to severe hearing losses at VIATAAL, KTH and HTCH. Speech intelligibility, perceived listening effort and overall preference (processed vs. unprocessed) of speech mixed with four different background noises are measured. So far, the German mild-hearing-loss user group has completed the gSASP measurements. In summary, these first results show that noise reduction hardly affects speech intelligibility and listening effort for this subject group. With respect to the overall preference, however, differences between processed and unprocessed conditions become more evident: one processed condition is very clearly, one slightly preferred; one condition is clearly, one slightly disliked. Evaluation of further conditions and research on partner communication support through the HaH platform continues.

Work package 7: Individual SASP Strategies

The iSASP first user studies have been performed in Oldenburg and show very reasonable results. All hearing impaired users experience an improvement after fitting the system in terms of speech intelligibility. But the different approaches show different strengths and therefore further work has to be done to combine the strengths of the different approaches.

Regarding the first user study results achieved with an early mockup of the iSASP wizard (as presented during the review meeting) further refinement steps were performed and the user studies for the whole test group were prepared. The wizard itself (as a result of the first mockup) is now modularized so that the fitting steps can be presented in a random order to improve the reliability of the results. Further the questions asked and interaction possibilities were improved, as there were quite frequent misunderstandings in the usage of the GUI for comparison step (“Play A”, “Play B”). This is now supported by better guided functions which directly show that only changes are possible during listening of “B”, as “A” is the reference audio. The accompanied questionnaire can be used as it was defined for the mockup as the test persons didn’t experience any problems filling them out and the questions cover the interesting topics.

The user tests then were prepared to take place at HTCH.. Tthe first final user studies started at the end of November and are completed by the end of December.. Additional young test users have been tested during January 2009. An additional “long-term” user test will be performed in May showing the potential of refitting the system during a longer period of usage (2-3 weeks)..

Work package 8: Home Automation Interfaces

The work in WP8 has progressed well developing the compromised software. Appropriate EHS & KNX/EIB dongles have been selected to connect both home automation networks to the HIC platform and the needed drivers and proxies to manage the networks have been developed and integrated with the HIC petri net based controller via the Home Automation Device Manager and the Home control Event Manager. Developments have been focus in reaching a flexible solution that helps integration of new devices and netoworks without having to rewrite the full controller if there show up new possible parameters and settings.

The GUI to allow the user interact with the system is the other component that has been developed within this WP. A mock-up of the GUI based on power-point screen-shots has been prepared to perform user tests in two different sites. A user-tests storyboard has been prepared, as well as a questionnaire to retrieve the information.