INTERSPEECH 2012 Speaker Trait Challenge

Whereas the first open comparative challenges in the field of paralinguistics targeted more "conventional" phenomena such as emotion, age, and gender, there still exists a multiplicity of not yet covered, but highly relevant speaker states and traits. In the last instalment, we focused on speaker states, namely sleepiness and intoxication. Consequently, we now want to focus on speaker traits: the INTERSPEECH 2012 Speaker Trait Challenge broadens the scope by addressing three less researched speaker traits: the computational analysis of personality, likability, and pathology in speech. Apart from intelligent and socially competent future agents and robots, main applications are found in the medical domain and surveillance.

For these Challenge tasks, the SPEAKER PERSONALITY CORPUS (SPC), the SPEAKER LIKABILITY DATABASE (SLD), and the NKI CCRT (Concomitant Chemo Radiation Treatment) SPEECH CORPUS (NCSC) with high diversity of speakers of different personality and likability and genuine pathologies will be provided by part of the organisers. The first – SPC - consists of 2 hours of French speech from 330 speakers labelled by 11 judges with standardised personality assessment tests, and will serve to evaluate features and algorithms for the estimation of speakers' personality traits in the popular "Big Five" OCEAN dimensions (openness, conscientiousness, extraversion, agreeableness, and neuroticism). The second – SLD – bases on the aGender corpus as employed in the INTERSPEECH 2010 Paralinguistic Challenge. Likability annotations by 32 labellers were added for 800 speakers in perfect age class and gender balance for roughly 1 hour of speech. Finally, NCSC provides 3 hours of Dutch speech from 40 speakers with head and neck cancer (tumours located in the vocal tract and larynx) recorded before and at various times after treatment. NCSC was created within the scope of an unrestricted research grant of Atos Medical, Sweden. The corpora feature further rich annotation such as speaker meta-data, orthographic transcript, phonemic transcript, and segmentation and multiple annotation tracks. All three are given with distinct definitions of test, development, and training partitions, incorporating speaker independence as needed in most real-life settings. Benchmark results of the most popular approaches will be provided as in the years before.

In these respects, the INTERSPEECH 2012 Speaker Trait Challenge shall help bridging the gap between excellent research on paralinguistic information in spoken language and low compatibility of results.

Three Sub-Challenges are addressed:

• In the Personality Sub-Challenge, the personality of a speaker has to be determined based on acoustics potentially including linguistics above or below average for the OCEAN five personality dimensions.
• In the Likability Sub-Challenge, the likability of a speaker's voice has to be determined by a suited learning algorithm and acoustic features. While the annotation provides likability in multiple levels, only two classes have to be recognised accordingly: likability above or below average.
• In the Pathology Sub-Challenge, the intelligibility of a speaker has to be determined by a suited classification algorithm and acoustic features.

The measures of competition will be Unweighted Accuracy and Area Under the receiver operating Curve (AUC). Transcription of the train and development sets will be known. All Sub-Challenges allow contributors to find their own features with their own machine learning algorithm. However, a standard feature set will be provided per corpus that may be used. Participants will have to stick to the definition of training, development, and test sets. They may report on results obtained on the development set, but have only five trials to upload their results on the test sets, whose labels are unknown to them. Each participation will be accompanied by a paper presenting the results that undergoes peer-review and has to be accepted for the conference in order to participate in the Challenge. The organisers preserve the right to re-evaluate the findings, but will not participate themselves in the Challenge. Participants are encouraged to compete in all Sub-Challenges.

Overall, contributions using the provided or an equivalent database are sought in (but not limited to) the following areas:

• Participation in the Personality Sub-Challenge
• Participation in the Likability Sub-Challenge
• Participation in the Pathology Sub-Challenge
• Novel features and algorithms for the analysis of speaker traits
• Unsupervised learning methods for speaker trait analysis
• Perception studies, additional annotation and feature analysis on the given sets
• Context exploitation in speaker trait assessment

The results of the Challenge will be presented at Interspeech 2012 in Portland, Oregon.
Prizes will be awarded to the Sub-Challenge winners.
If you are interested and planning to participate in the Speaker Trait Challenge, or if you want to be kept informed about the Challenge, please send the organisers an e-mail to indicate your interest.

To get started: Please obtain the License Agreement to get a password and further instructions for the download of the datasets: Please fill it out for the dataset(s) you wish to obtain, and sign and fax, accordingly.
After downloading the data you can directly start your experiments with the train and development sets. Once you found your best method you should write your paper for the Special Event. At the same time you can compute your results per instance and Sub-Challenge task on the test set and upload them. We will then let you know your according performance result.

Deadline:  01 April 2012

This is a event.