Overview

The so-called winter-over syndrome includes several detrimental symptoms such as insomnia, mild depression, irritability and social withdrawal (Suedfeld and Weiss, 2000). Hypoxic conditions at Concordia base represent an additional challenge with the possibility of the development of Acute Mountain Sickness and transient cognitive deficits. This is why Antarctic over-wintering is considered the best suited model for the study of adaptation to environmental challenge also pertinent in spaceflight conditions and particularly in long duration space missions.


In order to understand the process and dynamics of adaptation the effects should be continuously monitored. The need of frequent application of the tests over a long period of time poses a major challenge to the traditional psychological assessment. Language behaviour, however, is produced continuously; with the aid of language technology it can be analysed into indirect measures impervious to repetition (Balazs, 2007).


The present proposal aims at the analysis of language phenomena at multiple levels. Speech samples will be collected primarily in the form of semi-structured audio-video diaries. The collected speech samples will be analysed into measures of phonetics (articulation and prosody) as well as psychologically relevant indirect linguistic markers of verbal content.


The analyses will be performed with advanced methods of language technology. As much as current technology permits, automated analyses will be used allowing the highest degree of objectivity. By keeping manual contribution on the minimum, this proposal aims at providing a basis for the development of future fully automated systems that will allow using naturally occurring language manifestations by alleviating the burdens of confidentiality.


The acoustic-phonetic properties of speech are also relevant in the field of Medicine to some of the problems of physiological adaptation pertinent in isolation and particularly at Concordia; namely cerebral hypoxia and mild depression (SAD).

Phonetic data collection and processing

We aim at the examination of the sensitivity of acoustic-phonetic parameters of speech to hypoxia and to Seasonal Affective Disorder (SAD) and the development of a metric that alert crews at early stage of cognitive dysfunction (Automatic detection).


Speech data are collected from all crews using their mother tongue, signed up for the study weekly during their stay at the Concordia Station. Baseline data are collected in normal circumstances once before arrival as well as one month after departure from Antarctica. This way the inflections due to hypoxia and occasional occurrence of SAD symptoms are monitored. This collection of speech samples is called the Concordia Speech Database. Parallel with the data collection in the Concordia Station another data collection is also necessary from the seasonal affective disorder patients in normal atmospheric conditions, practically in a consulting room of the doctors. This, so called Seasonal Affective Disorder Database is necessary for the development of a good metric to detect SAD. Certainly, speech recording is necessary in symptomatic as well as in symptom free period of the same patient, for seeing the differences between the examined parameters. If for any reason this approach could not be pursued existing normal spontaneous speech databases (European Language resources Association Catalogue) could be used as references.


As the Concordia crew-members are French or Italian mother tongue, it is necessary to collect SAD database for both languages, because it may be assumed, that there are difference in the measures of phonetics, prosody, and content, according to the language. Moreover the seasonal affective disorder database collection will be prepared for three languages: Italian, French, and Hungarian. This way, it is possible to examine the language dependency or independency of the seasonal affective disorder inflection on acoustic-phonetic parameters of speech. Hungarian language is added, for the examination of language dependency in a language that belongs to a different language family.


The speech databases are composed of two parts: the first is spontaneous speech obtained from the voice diaries used for content analysis, The second is reading a text, a standard phonetically balanced short folk tale (about 6 sentences all together), “The North Wind and the Sun”. This story is from the booklet "The Principles of the International Phonetic Association" (International Phonetic Association, 1999). In this booklet phonetic transcriptions of the story are given in 50-odd languages, in order to illustrate the use of the International Phonetic Alphabet (IPA). The tale is frequently used in the phoniatric practice for all European languages, and commonly used for acoustic analysis in the framework of international, multi-lingual clinical studies.


The seasonal affective disorder database gives possibilities of the selection of the optimal acoustic phonetic parameters which characterize the SAD.


Statistical analyses are aided by the artificial speech recognition techniques developed in the Laboratory of Speech Acoustics (Vicsi and Szaszák, 2010; Vicsi and Szaszák, 2005). For example for the selection of speech sounds, and groups of them, automatic segmentation and labeling will be used.


The database collection in the Concordia station allows us to examine the inflection of the hypoxia on the speech parameters, and to follow how the mood of the patient is changing according to time spent at the station.

Concordia Speech Database collection

Baseline speech samples will be obtained from the crews before the expedition, at one occasion, in normal circumstances,. The following tasks must be recorded from each crew member:


1) speaking a daily voice diary about 5 min.


2) reading the standard phonetically balanced short folk tale (about 6 sentences all together), frequently used in the phoniatry practice for all European languages, “The North Wind and the Sun”;


During the expedition, two types of personal speech data will be recorded from all crews:


Recording once a week: daily voice diaries, about 5 min.


Recording once a month: the phonetically balanced short folk tale “The North Wind and the Sun”;


Recording conditions: using near-field microphone (Monacor ECM-100), with Creative Soundblaster Audigy 2 NX: an external USB sound card at a 16 kHz sampling rate, quantized at 16 bits.

Seasonal Affective Disorder Speech Database collection

Speech data collection is also necessary from the seasonal affective disorder patients in normal atmospheric conditions, practically in a doctor’s consulting room. The seasonal affective disorder database collection is planned to prepare for three languages Italian, French, and Hungarian. Personal speech recording as described in Concordia Speech Database collection above are performed.

Acoustic phonetic analyses and classification

The following acoustic phonetic parameters and their statistics are measured

  • pitch frequency f0, jitter,
  • level of voice intensity, shimmer,
  • Voice Onset Time (VOT),
  • spectrum, formant values, noise-to-harmonic component ratio
  • durations of different speech units, speech rate, rhythm, articulation speed
  • change of pitch in different linguistic units as in words, clauses, sentences
  • percent time pausing, vocalization/pause ratio
Parameters of phonation and articulation are measured in speech sound as basic units while prosodic parameters are measured in longer units such as words, clauses or sentence. For the automatic selection of the different speech units automatic speech recognizer will be used separately for Italian, French and Hungarian languages. It is very important to find the most characteristic parameters for hypoxia and SAD separately. For the selection of the optimal set of parameters factor analyses are performed. On the basis of artificial intelligence techniques, classification experiments are developed to automatically classify healthy and inflected speech by hypoxia and by SAD. Super Vector Classification technic (SVM), especially the Least Square- SVM is a best practice in the Laboratory of Speech Acoustics, but we use other technics too, as fuzzy decision-making systems.