Conversation Piece

  Alexa Wright + Alf Linney, 2008-9

   
 


Conversation Piece is an interactive computer installation that mimics social relations with human users. In the installation a disembodied synthesized voice will try to engage individual audience members in dialogue. Each interaction is focused around a small sculpture displayed on an exhibition plinth. People entering the space are automatically tracked using webcams positioned overhead. When someone moves past one of the sculptures the disembodied synthesized voice of our artificial agent, ‘Heather’ tries to catch his or her attention by saying ‘Hello’, or ‘Excuse me’. As an individual approaches one of the sculptures ‘Heather’ will start a conversation. Using keywords to interpret what is said in reply, she will pursue a dialogue with the user that can be heard only at one particular location in the space.


 

 
 


We use speech recognition and synthesis software, concealed microphone arrays, a dialogue management system and directional sound sources to enable dialogues to be conducted simultaneously at two different locations. Rather than relying on traditional input devices to facilitate interactivity, the interface is rendered transparent by the use of concealed microphone arrays and focused directional sound. The directional sound sources ensure that ‘Heather's’reactions can only be heard by someone standing close to a plinth. Absorption panels are used to enhance the directionality of the sound. It is important that the technology is hidden because this helps to give the illusion that the virtual character really is listening and responding. It encourages people interacting with the work to attribute human sensibilities to the machine, even in the absence of any visible human features.

 
 

 
 



 
   
 


Technical Notes:
The installation takes approximately 3 days to set up.
The system has an auto-restart so that it will be able to recover from most crash situations during exhibition and should therefore require little maintenance. It is programmed to restart automatically every day at 10am.

Ideally two ‘talking points’ require a space of around 8m x 10m, but this is negotiable and if space is restricted one ‘talking point’ could be exhibited.

The installation requires constant, flat and reasonably bright lighting – fluorescent usually works well. We need good access to power points (see diagram).

The installation preferably needs its own space, but if this is not possible background noise must be minimal. This is very important as the voice cannot be heard in a very noisy space (the ultrasound waves cannot be amplified), and also the word recognizer is confused by loud background sounds. The directional speakers work best in a large space, so the two talking points can be spaced well apart, but this is not essential.