Activities

No upcoming events

Teleservices for an Intelligent Room

Participants

• Timo Dankert
• Sabine Klein
• Kai Mismahl
• Alexander Schulz

Supervisor

• Thies Pfeiffer

Motivation

• Users often face the problem to be unable to communicate with a room in an intuitive way
• In normal room environments users have to use a keyboard to give commands
• An Intelligent Room could help to solve this problem by introducing itself and listening to the user's voice

Application Szenario

A typical situation would be a user who enters the intelligent room and is welcomed by the room itself. The user is able to communicate with the room in an intuitive way through speech. Commands will be recognized by the system and executed.

Objectives

The project goals are

• Recognition of speech in German
• Execution of commands
• Text to speech system
• Integration of other projects

Description

The system is using Esmeralda for speech recognition, MaryTTS for text to speech and JVoiceXML for the dialogue control. For the communication we chose XCF.
Esmeralda recognizes German words and sentences, which are defined in a grammar. The input is forwarded to JVoiceXML, where the dialogue itself is performed. By using VoiceXML files it is possible to parse the input and reply to the input either with text output or calling JavaMethods to perform different actions.
The output is forwarded via XCF to MaryTTS to create spoken output by a synthesized voice.

Results

The system is running and has been demoed. Its features are:

• German speech recognition
• Dialogue management with VoiceXML files
• German text to speech
• Integrated project (Hausgeist)

Demonstration video
The video shows a user who is performing a short dialogue with the system. By giving commands he is able to switch on and off the light and music. With the command 'Hausgeist Tour starten' the system is starting another project for a short tour through the Intelligent Room. The other project uses MaryTTS, too.

Discussion and Conclusion

• All in all, the project goal was achieved in a satisfactory way
• The dialogue flow was designed with the help of dialogue analysis and collected user data
• Still, the speech recognition causes some difficulties:
• The recognition success is very speaker sensitive
• Recognition of more complex speech constructs is still not satisfactory

Outlook

It is very easy to improve the system by using VoiceXML files. Nevertheless there are some points which could be improved:

• Better speech recognition
• New dialogues
• More integrated projects
• A different TTS system to give personality to the system