A synthesis-by-rule system based on the selective use of non-uniform synthesis units has been developed. This system uses a natural speech database and an algorithm which searches the database for the optimal speech segment to be used as the synthesis unit. Because of flexible use of synthesis units, this scheme has great advantages, especially in expressing many coarticulat~ry variations. Howe...

In this paper, we discuss two distinct data sets. The first relates to the so-called allophonic process of closed-syllable laxing in Québec French, which targets final (stressed) vowels even though these vowels are arguably syllabified in open syllables in lexical representations. The second is found in the forms produced by a first language learner of European French, who displays an asymmetry...

Canadian Raising is the best-known of a diverse class of English allophonic height alternations in /ai/ conditioned by coda voicing. The alternations have been independently re-innovated and show a systematic typology: The voiceless environment selects the higher allophone. We hypothesize that the phonetic basis of the asymmetry is the tendency for diphthongs to assimilate to their nuclei befor...

The distribution of fine-grained phonetic variation can be observed in the speech of members of welldefined social groups. It is evident that such variation must somehow be able to propagate through a speech community from speaker to hearer. However, technological barriers have meant that close and direct study of the articulatory links of this speaker-hearer chain has not, to date, been possib...

In the context of large vocabulary speech recognition systems it is of major importance to accurately model the allophonic variations to be faced in a real world task. Evaluation of which variants are actually improving the system performance is crucial, as it determines the acceptance of the pronunciation alternatives used. Traditional approaches in this direction use different criteria and, t...

The distribution of the flap allophone [n] of American English, along with the other allophones of /t/, [t5, t , …, t] has been accounted for in various formal frameworks by = assuming a number of different abstract mechanisms and entities. The desirability or usefulness of these formalisms is not at issue in the present paper. Instead, a computationally explicit model of categorization is used...

Intelligible Text-to-Speech may be achieved by concatenating spectrally encoded segments. However, its Iack of naturalness could be attributed to a difficult control of speech parameters. Acoustic rules are more adequate for this control. The aim of this work is to provide a methodology to move from a segmental to a rule-based approach. A number of interactive tools is proposed using powerful s...

This paper argues that training Generative Adversarial Networks (GANs) on local and non-local dependencies in speech data offers insights into how deep neural networks discretize continuous symbolic-like rule-based morphophonological processes emerge a convolutional architecture. Acquisition of has recently been modeled as dependency between latent space generated by GANs Beguš (2020b), who mod...