This paper presents two case studies that highlight both the challenges and benefits of working with large-scale, naturalistic phonetic data. Our aim is to encourage researchers not to shy away from phonetic data found “in the wild”, even when such data are messy, noisy, or incomplete – because they can yield robust, novel insights beyond the reach of controlled laboratory studies. We focus on challenges that are endemic to large corpora, including degraded audio quality, sparse or inconsistent annotations, and missing speaker metadata. By comparing two corpus-based studies that diverge in methodology and statistical design, we show how different approaches can mitigate these limitations while still extracting meaningful patterns.

Our native language influences the way we perceive speech sounds, affecting our ability to discriminate non-native sounds. We compare two ideas about the influence of the native language on speech perception: the Perceptual Assimilation Model, which appeals to a mental classification of sounds into native phoneme categories, versus the idea that rich, fine-grained phonetic representations tuned to the statistics of the native language, are sufficient. We operationalise this idea using representations from two state-of-the-art speech models, a Dirichlet process Gaussian mixture model and the more recent wav2vec 2.0 model. We present a new, open dataset of French- and English-speaking participants’ speech perception behaviour for 61 vowel sounds from six languages. We show that phoneme assimilation is a better predictor than fine-grained phonetic modelling, both for the discrimination behaviour as a whole, and for predicting differences in discriminability associated with differences in native language background. We also show that wav2vec 2.0, while not good at capturing the effects of native language on speech perception, is complementary to information about native phoneme assimilation, and provides a good model of low-level phonetic representations, supporting the idea that both categorical and fine-grained perception are used during speech perception.

L‟article traite de la manifestation acoustique des consonnes latérales en anglais américain et en roumain en fonction de la position syllabique et de la complexité phonotactique. Nous avons considéré quatre types de mesures: valeurs formantiques, équations locus, ratio d‟intensité et présence/absence de relâchements. Notre but est, d‟une part, de classifier les allophones des deux langues considérées et d‟autre part de déterminer les indices acoustiques des gestes articulatoires des consonnes latérales. Les résultats indiquent des différences importantes entre les deux langues. On montre que la distribution des allophones n‟est pas binaire, mais graduée et que le statut du geste dorsal peut être considéré comme un marqueur de « degré de clarté ». On montre aussi que l‟allophonie dépend de la position syllabique mais pas forcément de la complexité syllabique.

We address the task of stress prediction as a sequence tagging problem. We present sequential models with averaged perceptron training for learning primary stress in Romanian words. We use character n-grams and syllable n-grams as features and we account for the consonant-vowel structure of the words. We show in this paper that Romanian stress is predictable, though not deterministic, by using data-driven machine learning techniques.