Project #06.1: Laughter Cycle (Query-by-Example and Navigation Approaches for Audio Content)
This project aims at developing a web-based service for efficiently navigating inside a laughter database. It will enable users to quickly and easily find laughter utterances corresponding to their wishes. Two main approaches to explore the database will be provided: query-by-example (the user gives a laughter utterance to the system, which finds similar samples in the database) and query-by-keyword (using tags stored as meta-data and characterizing the utterances).
The query-by-example approach requires an evaluation of similarities between laughter occurrences. Several acoustic properties will be investigated to assess to what extent they model human perceptions of similarities between laughters. Characteristics usually computed in speech and music analysis (e.g. descriptors of pitch, timbre, harmony, melody or rhythm) will be adapted to laughter. Features specific to laughter will also be considered, for example the ratio of voiced versus unvoiced segments or descriptors characterizing the periodicity and durations of such segments. The laughter utterances from the database will be visually organized according to their similarity with the example along one dimension and the similarities between them along another space dimension, in order to ease the localization of the region of interest. The user will be able to listen to any displayed sample, define it as the new example towards which similarities must be estimated and reorganize the database visualization by, for example, weighting the acoustic properties involved in the similarity evaluation.
The query-by-keyword approach will only differ in the way data is accessed. According to the requested keyword(s) and the database tags (possibly automatically extracted through characteristic features aforementioned), the software will propose the most relevant samples and place all the objects around them. The user will then be able to navigate inside the organized database, using either keywords or acoustic similarities to refine his search.
The software will be embedded in a web-based service. This implies providing a quick access to the laughter database remotely stored and managed on one hand and, on the other hand, developing a web application to run the audiovisual rendering on the client computer. In addition, this application must enable the user to record its own laughter and send it to the server for starting a query-by-example search. In consequence, world-wide users will be able to enrich the database with new laughter utterances and find similarities with other laughers. The website will support a laughing community. Designing this website, starting by establishing its specifications and studying solutions to develop it (Rich Internet Applications being good candidates), will be an important part of the project. It can be split into 3 aspects corresponding to 3 developers profiles:
- server side application, managing the database of laughters and community members, likely based on PHP/MySQL or some equivalent technologies.
- client side GUI and interaction, probably using Flash or AJAX.
- client side graphics design.
Laughter is an essential signal in human communications. It conveys information about our feelings and helps to cheer up our mood. Moreover, it is communicative and has the potential to elicit emotions to its listeners. Laughter is also known to have healthy effects, and especially to be one of the best medicines against stress. Laughter therapies, “yoga” sessions or groups are emerging everywhere, amplified by these crisis times where people need to relax and forget about the negative things. We can cite the laughter chain launched on Skype, laughters clubs or the World Laughter Day. The first objective of Laughter Cycle is concerned by this strong social impact. Inspired by singing communities like Midomi and the miscellaneous social networks emerging over the internet, we would like to create a laughing community, where users could enjoy all the benefits of laughter.
In addition, due to the growing interest for virtual agents modeling human behaviors, there is a need to enable these agents to perceive and express emotions. Laughter is clearly an important cue for understanding emotions and discourse events on one hand, and, on the other, to manifest certain emotions. However, from an engineering point of view, laughter is an understudied signal. The second goal of this project will be to extend the state-of-the art in laughter production. Pure laughter synthesis is very difficult and, to date, unefficient. We would like to help people finding easily a desired laughter sound. We have acquired 2 databases containing several thousands of laughters, covering the largest possible range of laughter kinds, that will be primarily used to this end. Organizing the database according to similarities will facilitate and accelerate the search.
This project lies in the continuity of the work done within a previous numediart project Audio Cycle (#04.1), where a graphical interface to browse a large music library and help creating new performances has been designed. Similarities between music loops have been used to spatially organize the corpus and synchronized 3D Audio rendering of the activated loops has been implemented.