Alcohol Language Corpus Research Articles

Beyond breathalyzers: AI-powered speech analysis for alcohol intoxication detection

Detecting potential alcohol inebriation or intoxication status holds paramount significance for social prevention and security. Beyond its association with long-term health effects, alcohol consumption can lead to immediate consequences, including reduced control over one’s actions, with traffic fatalities representing one of the most tragic outcomes.This study leveraged the Alcohol Language corpus, involving 162 subjects recorded both in sober and inebriated states. Participants provided 60 speech samples while sober and 30 when intoxicated, all within a realistic car setting using head-mounted microphones. Our research endeavors encompassed comprehensive stratified statistical tests to examine the impact of alcohol consumption on speech production while uncovering the influence of covariates such as age, gender, and drinking habits.Additionally, we introduced a speaker-neutral machine learning algorithm, based on the Domain-Adversarial Neural Network architecture. This approach aimed to overcome challenges posed by individual differences that often complicate intoxicated speech analysis. Notably, our findings highlighted the effectiveness of features like the RASTA-filtered auditory spectrum. Nevertheless, the results from statistical tests emphasized the need for techniques that minimize inter-subject variability.As for the automatic classification, the proposed architecture exhibited promising results, yielding a classification accuracy slightly exceeding 70% on an independent test set. Although preliminary, our research demonstrates the potential for detecting alcohol-induced speech changes, benefiting societal well-being and security. It also underscores the importance of developing strategies that account for individual differences while harnessing the power of automatic models to effectively distinguish between sober and intoxicated individuals.

Vowel space expansion following alcohol intoxication

Alcohol intoxication is characterized by hypermetric movements (overshoot) in manual fine motor skills [Phillips et al., (2009). Hum. Movement Sci,. 28(5), 619–632]. It remains unknown whether hypermetric movements manifest in speech following alcohol intoxication. Recent work on a small population (n = 15) revealed vowel space area (VSA) expansion following intoxication [Chang et al., 2023. Proc. ICPhS ‘23 ]. To validate these findings, we made use of the publicly available Alcohol Language Corpus [Schiel et al., 2012. Lang. res. and eval. 46, 503–521]. VSAs were compared across 162 speakers (85 male, 77 female) while sober (n = 162) and intoxicated (blood alcohol level > 0.08%, n = 97) using an unsupervised method based on cluster center detection of normalized vowel formants [Sandoval et al. JASA. 134(5), EL477-EL483]. Substantial VSA expansion was observed for males (10.8%), while a smaller expansion was observed for females (3.6%). These results support recent observations of VSA expansion and suggest hypermetric tongue movement following alcohol intoxication. Further evaluations are ongoing and employ hierarchical linear modeling to compare the effects of specific blood-alcohol concentrations and speech tasks (free speech, tongue-twisters, repetition) on VSA expansion. [Research funded by Tenvos Incorporated for the development of commercial speaker state-detection algorithms.]

Voice analysis for intoxication detection in laboratory versus law enforcement contexts

There is substantial interest in deploying voice biomarkers for the detection of alcohol intoxication, yet it remains unknown how other mental and physical states (e.g. emotion, stress) influence voice biomarkers in intoxicated speech. We compared measurements of voice quality (jitter, shimmer, noise-harmonics ratio) across two datasets of alcohol-intoxicated speech: (1) The alcohol language corpus, which contains laboratory elicited speech from 167 individuals in both sober and intoxicated conditions and (2) a custom dataset of police (control, n = 14) and suspect (intoxicated, n = 32) interactions during traffic stops where intoxication was verified via breath analysis. Measurements were extracted from all stressed vowel tokens and compared across conditions using two-sample t-tests within sex-specific groupings of each dataset. For both males and females, jitter was significantly lower during intoxication as measured from laboratory-elicited speech, but significantly higher for intoxicated individuals when measured from police–suspect interactions. These results suggest that voice biomarkers for alcohol-intoxication are easily confounded by other emotional and physical states (e.g., stress during police interaction), and thus, present a particular challenge for speaker-independent detection systems where baseline voice quality measurements across different emotional states are unknown. [Research funded by Tenvos Incorporated for the development of commercial speaker state-detection algorithms.]

Feature generalizability for speaker-dependent detection of alcohol intoxication

Impairments to speech motor control from alcohol intoxication are variable across individuals, making speaker-dependent approaches ideal for speech-based intoxication detection [Schiel et al., 2010. Proc. INTERSPEECH 2010]. Here, we evaluated whether individual acoustic features have high generalizability across speaker-dependent models. We selected 97 speakers (54 male, 43 female) from the Alcohol Language Corpus [Schiel et al., 2012. LRE. 46, 503-521] who had sufficient sober and intoxicated (>0.08% blood-alcohol concentration) recordings for speaker-dependent modeling. For each speaker, we extracted 9 features from vowels (F0–F3, jitter, shimmer, harmonics-to-noise ratio, duration, and duration variability) and 7 from consonants (spectral skewness and kurtosis, center of gravity, duration and duration variability, harmonics-to-noise ratio), and fitted these to speaker-dependent random forest models with 5-fold cross-validation to evaluate feature importance from the associated mean decrease in Gini impurity (GI). Across all speakers, consonant-based features tended to have stronger generalizability than vowel-based features, with spectral skewness and kurtosis being the most generalizable (GI: 0.11 and 0.09), and vowel duration and F2 being the least generalizable (GI: 0.04 and 0.03). Further experiments to explore additional features and evaluate sex-specific generalizability are ongoing. [Research funded by Tenvos Incorporated for the development of commercial speaker state-detection algorithms.]

Audio-based Deep Learning Algorithm to Identify Alcohol Inebriation (ADLAIA).

Acute alcohol intoxication impairs cognitive and psychomotor abilities leading to various public health hazards such as road traffic accidents and alcohol-related violence. Intoxicated individuals are usually identified by measuring their blood alcohol concentration (BAC) using breathalysers that are expensive and labour-intensive. In this paper, we developed the Audio-based Deep Learning Algorithm to Identify Alcohol Inebriation (ADLAIA) that can instantly predict an individual's intoxication status based on a 12-second recording of their speech. ADLAIA was trained on a publicly available German Alcohol Language Corpus that comprises a total of 12,360 audio clips of inebriated and sober speakers (total of 162, aged 21-64, 47.7% female). ADLAIA's performance was determined by computing the unweighted average recall (UAR) and accuracy of inebriation prediction. ADLAIA was able to identify inebriated speakers-with BAC of 0.05% or higher-with an UAR of 68.09% and accuracy of 67.67%. ADLAIA had a higher performance (UAR of 75.7%) in identifying intoxicated speakers (BAC > 0.12%). Being able to identify intoxicated individuals solely based on their speech, ADLAIA could be integrated in mobile applications and used in environments (such as bars, sports stadiums) to get instantaneous results about inebriation status of individuals.

ОБ ОЦЕНКЕ ВЛИЯНИЯ АЛКОГОЛЬНОЙ ИНТОКСИКАЦИИ НА ПАРАМЕТРЫ ГОЛОСА В СИСТЕМАХ БИОМЕТРИЧЕСКОЙ АУТЕНТИФИКАЦИИ

Changing the voice characteristics of a person under the influence of alcohol intoxication affects the effectiveness of the identification procedure. From which it follows that the determination of the level of alcohol in the blood remotely is important for ensuring information security in computer systems. The purpose of the present paper is to analyze the existing voice authentication methods and approaches to improve accuracy, analyze existing data sets, and evaluate the influence of the psychophysiological state of the speaker on the parameters of his speech. The focus of this study is on exploring different approaches to recognizing alcohol intoxication and evaluating their effectiveness. The study includes a comparative analysis of these methods, emphasizing the importance of using a comprehensive and representative data set that takes into account factors such as gender, age, stage of intoxication, recording quality, and ambient noise levels. Based on the analysis of scientific publications, the study identifies the random forest method as one of the most effective machine learning methods, demonstrating an accuracy of 95.3% for all speech datasets and 80% for the widely used Alcohol Language Corpus.

An efficient state detection of a person by fusion of acoustic and alcoholic features using various classification algorithms

In this paper, we present a novel approach for extracting emotional information along with the state of intoxication. Conventional methods use features for identifying the state of intoxication or emotion recognition distinctly. In this work, we extract the efficient features from Alcohol Language Corpus for alcohol state detection and Berlin dataset for emotional behavior. The features extracted are fused after the feature extraction. Through the proposed approach, we can extract information of the driver, such as drunken state and emotional state at the same time. The paper deals with the driver state classification, whether he/she is alcoholic, non-alcoholic, and also their emotional behavior, such as happy, anger, sad, fear, neutral, etc. from speech signals. The main application of the work is to safeguard a person’s life who is a daily user of vehicles and alert him from accidental prone situations. We have used classifiers such as Support Vector Machine, K Nearest Neighbor, Random Forest, Gradient Boosting, and Extremely Randomized Trees. The outcome is to detect the emotion and intoxication state of the driver.

The influence of alcoholic intoxication on the short-time energy function of speech.

This study investigates rhythmic features based on the short-time energy function of speech signals with the aim of finding robust, speaker-independent features that indicate speaker intoxication. Data from the German Alcohol Language Corpus, which comprises read, spontaneous, and command&control speech uttered by 162 speakers of both genders and various age groups when sober and intoxicated, were analyzed. Energy contours are compared directly (Root Mean Squared Error, statistical correlation, or the Euclidean distance in the spectral space of the contour) and by parameterization of the contour using the Discrete Cosine Transform (DCT) and the first and second moments of the lower DCT spectrum. Contours are also analyzed by Principal Components Analysis aiming at fundamental "eigen contour" changes that might encode intoxication. Energy contours differ significantly with intoxication in terms of distance measures, the second and fourth DCT coefficients, and the first and second moments of the lower DCT spectrum. Principal Components Analysis did not yield interpretable "eigen contours" that could be used in distinguishing intoxicated from sober contours.

Intoxicated speech detection: A fusion framework with speaker-normalized hierarchical functionals and GMM supervectors

Segmental and suprasegmental speech signal modulations offer information about paralinguistic content such as affect, age and gender, pathology, and speaker state. Speaker state encompasses medium-term, temporary physiological phenomena influenced by internal or external bio-chemical actions (e.g., sleepiness, alcohol intoxication). Perceptual and computational research indicates that detecting speaker state from speech is a challenging task. In this paper, we present a system constructed with multiple representations of prosodic and spectral features that provided the best result at the Intoxication Subchallenge of Interspeech 2011 on the Alcohol Language Corpus. We discuss the details of each classifier and show that fusion improves performance. We additionally address the question of how best to construct a speaker state detection system in terms of robust and practical marginalization of associated variability such as through modeling speakers, utterance type, gender, and utterance length. As is the case in human perception, speaker normalization provides significant improvements to our system. We show that a held-out set of baseline (sober) data can be used to achieve comparable gains to other speaker normalization techniques. Our fused frame-level statistic-functional systems, fused GMM systems, and final combined system achieve unweighted average recalls (UARs) of 69.7%, 65.1%, and 68.8%, respectively, on the test set. More consistent numbers compared to development set results occur with matched-prompt training, where the UARs are 70.4%, 66.2%, and 71.4%, respectively. The combined system improves over the Challenge baseline by 5.5% absolute (8.4% relative), also improving upon our previously best result.

The influence of alcoholic intoxication on the fundamental frequency of female and male speakers

This study investigates long-term features and utterance contours of fundamental frequency (f0) derived from the German Alcohol Language Corpus. The corpus comprises read, spontaneous, and command&control speech uttered by 148 speakers of both genders and various age groups when sober and intoxicated. f0 median, f0 range, and f0 contours are analyzed for intoxication and interactions with gender and age. Contours are compared both directly (root mean squared error, statistical correlation, or the Euclidean distance in the spectral space of the contour) and by parameterization of the contour using discrete cosine transform and the first and second moment of the lower contour spectrum. Results partly confirm earlier findings, i.e., f0 average and range are mostly raised with intoxication, and also suggest that the majority of speakers do not follow a general trend, but show idiosyncratic alterations to f0. f0 contours differ significantly with intoxication, but a more detailed analysis could not assign these changes to specific general form changes like decline or curvature. The results suggest that it is not possible to predict intoxication from f0 in a single model across different speakers. Instead a speaker-dependent model to account for the individual speaker behavior is proposed.

Alcohol language corpus: the first public corpus of alcoholized German speech

The Alcohol Language Corpus (ALC) is the first publicly available speech corpus comprising intoxicated and sober speech of 162 female and male German speakers. Recordings are done in the automotive environment to allow for the development of automatic alcohol detection and to ensure a consistent acoustic environment for the alcoholized and the sober recording. The recorded speech covers a variety of contents and speech styles. Breath and blood alcohol concentration measurements are provided for all speakers. A transcription according to SpeechDat/Verbmobil standards and disfluency tagging as well as an automatic phonetic segmentation are part of the corpus. An Emu version of ALC allows easy access to basic speech parameters as well as the us of R for statistical analysis of selected parts of ALC. ALC is available without restriction for scientific or commercial use at the Bavarian Archive for Speech Signals.