Voiceless Codas Research Articles

German Word-Final Devoicing in Naturally-Produced and TTS Speech

This study explores the production and perception of word-final devoicing in German across text-to-speech (from technology used in common voice-AI “smart” speaker devices—specifically, voices from Apple and Amazon) and naturally produced utterances. First, the phonetic realization of word-final devoicing in German across text-to-speech (TTS) and naturally produced word productions was compared. Acoustic analyses reveal that the presence of cues to a word-final voicing contrast varied across speech types. Naturally produced words with phonologically voiced codas contain partial voicing, as well as longer vowels than words with voiceless codas. However, these distinctions are not present in TTS speech. Next, German listeners completed a forced-choice identification task, in which they heard the words and made coda consonant categorizations, in order to examine the intelligibility consequences of the word-final devoicing patterns across speech types. Intended coda identifications are higher for the naturally produced productions than for TTS. Moreover, listeners systematically misidentified voiced codas as voiceless in TTS words. Overall, this study extends previous literature on speech intelligibility at the intersection of speech synthesis and contrast neutralization. TTS voices tend to neutralize salient phonetic cues present in natural speech. Subsequently, listeners are less able to identify phonological distinctions in TTS. We also discuss how investigating which cues are more salient in natural speech can be beneficial in synthetic speech generation to make them more natural and also easier to perceive.

上古漢語聲調的起源

In this papaer, I researched on Old-Chinese tones around three mainstream theories in academia. There are "The Voiced codas and Voiceless codas hypothesis", "The Long vowels and Short vowels hypothesis" and “The codas hypothesis”.
 I believe that "The codas hypothesis" is more persuasive among the three theories. Because "The codas hypothesis" can not only explain the internal problems of Old-Chinese, but also explain the correspondence between the Sino-Tibetan languages such as Written Tibetan and Written Burmese, the loanwords of the neighboring languages such as Korean and Japanese, the translated materials such as Sanskrit-Chinese.
 So I think that the departing tone(去聲) is derived from -s in Old-Chinese. This conclusion is consistent with other scholars who advocate “The codas hypothesis”.
 On the other hand, most scholars who advocate "The codas hypothesis" believe that rising tone(上聲) is derived form -ʔ. But I think the rising tone(上聲) is derived from -q and -ɦ in Trans Himalayan(Proto-Sino-Tibetan). Because this point of view can better explain the internal and external problems of Old-Chinese. However, the codas -q and-ɦ in Trans Himalayan(Proto- Sino-Tibetan), have been merged into -q in Old Chinese.

Syllable-wide distribution of acoustic cues to coda voicing

In English, the strongest acoustic cues to preserving the voicing contrast in coda stops are in the preceding vowel: a voiced coda is associated with a longer vowel and a voiceless coda with a shorter vowel. Our recent work (Jacewicz et al., 2021) examining structured variability in stop voicing implementation in female productions showed that the cues to stop coda voicing extended to the syllable initial stop. In running speech, the /b/-closure in “bad” was shorter when coda was voiced, and it was longer when coda was voiceless (“bat”). Here, we test the hypothesis that voicing contrast cueing the “bad-bat” distinction is reinforced syllable-wide and involves specific long-distance timing relationships between closures of both stops, the extent of their closure voicing, vowel duration, and positive VOT. The current dataset consists of 2610 productions by 45 adult males, who are also diversified by dialect. Preliminary analyses confirmed that segmental voicing information is distributed over long domains (here, a monosyllabic word) and that cues to coda voicing are available in the syllable onset. These findings imply that temporal relationships among acoustic phonetic detail cueing lexical distinctions can potentially enhance the perceptual dimensions of perceived syllable- or word-wide voicelessness and voicing.

Variability in within-category implementation of stop consonant voicing in American English-speaking children.

The development of stop consonant voicing in English-speaking children has been documented as a progressive mastery of phonological contrast, but implementation of voicing within one voicing category has not been systematically examined. This study provides a comprehensive account of structured variability in phonetic realization of /b/ in running speech by 8-12-year-old American children (n = 48) when compared to adults (n = 36). The stop always occurred word-initially, was followed by either a voiced or voiceless coda, and its position varied in a sentence, which created systematic conditions to examine acoustic variability in closure duration (CD) and voicing during the closure (VDC) stemming from phonetic context and prosodic prominence. Children demonstrated command of long-distance anticipatory coarticulation, providing evidence that information about coda voicing is distributed over an entire monosyllabic word and is available in the onset stop. They also manifested covariation of cues to stop voicing and command of prosodic variation, despite greater random variability, greater CD, reduced VDC, and exaggerated execution of sentential focus when compared to adults. Controlling for regional variation, dialect was a significant predictor for adults but not for children, who no longer adhered to the marked local variants in their implementation of stop voicing.

The Acquisition of Acoustic Cues to Onset and Coda Voicing Contrasts by Preschoolers With Hearing Loss.

Children with hearing loss (HL), including those with hearing aids (HAs) and cochlear implants (CIs), often have difficulties contrasting words like " b each" versus " p each" and "do g " versus "do ck " due to challenges producing systematic voicing contrasts. Even when acoustic contrasts are present, these may not be perceived as such by others. This can cause miscommunication, leading to poor self-esteem and social isolation. Acoustic evidence is therefore needed to determine if these children have established distinct voicing categories before entering school and if misperceptions are due to a lack of phonological representations or due to a still-maturing implementation system. The findings should help inform more effective early intervention. Participants included 14 children with HL (eight HA users, five CI users, and one bimodal) and 20 with normal hearing, all English-speaking preschoolers. In an elicited imitation task, they produced consonant-vowel-consonant minimal pair words that contrasted voicing in word-initial (onset) or word-final (coda) position at all three places of articulation (PoAs). Overall, children with HL showed acoustically distinct voicing categories for both onsets and codas at all three PoAs. Contrasts were less systematic for codas than for onsets, as also confirmed by adults' perceptual ratings. Preschoolers with HL produce acoustic differences for voiced versus voiceless onsets and codas, indicating distinct phonological representations for both. Nonetheless, codas were less accurately perceived by adult raters, especially when produced by CI users. This suggests a protracted development of the phonetic implementation of codas, where CI users, in particular, may benefit from targeted intervention.

A perceptual pathway for voicing-conditioned vowel duration

When codas and vowels are cross-spliced, vowels originally produced with voiced codas are perceived as longer than vowels of the same duration produced with voiceless codas. The spliced coda has the opposite effect: Vowels presented with voiced codas are perceived as shorter. To explain what characteristics make vowels produced with voiced codas sound longer than vowels produced with voiceless codas, four experiments tested how acoustic correlates of voicing affect English speakers’ perception of vowel duration. Vowels were manipulated in a ten-step duration continuum, and listeners categorized each vowel as ‘long’ or ‘short.’ Study 1 tested effects of vowel height categories (/æ, ε, ɪ/) and within-category F1. Study 2 tested effects of intensity contour. Study 3 tested effects of spectral tilt. Perceived vowel duration increased with vowel height and rising intensity. Perceived vowel duration decreased with falling intensity and with higher spectral tilt. There was no effect of within-category F1. Study 4 confirmed that the effect of the original coda environment is not specific to English stimuli. The effects of spectral tilt and intensity contour on perceived duration provide a possible perceptual pathway for the development of voicing-conditioned vowel duration; coda voicing influences these characteristics and they in turn influence perceived vowel duration.

Influence of coda stop features on perceived vowel duration

Four experiments tested what cues contribute to English speakers’ perception of vowel duration. Listeners categorized the duration of vowels as ‘long’ or ‘short’ for stimuli produced with voiced, voiceless, breathy voiced, or voiceless aspirated stop codas. Listeners demonstrated a strong ability to perceive vowel duration, though perception was continuous rather than categorical. There were several interacting factors influencing perceived vowel duration, based on expectations set by the presence of particular codas and also acoustic effects of the coda on the vowel. When the coda was removed, vowels that had been produced before voiced codas were perceived as longer than vowels produced before voiceless codas, though they exhibited the opposite effect when codas were present. Vowels were also perceived as longer when produced before breathy voiced stops, regardless of whether or not the stop was present. The steeper f0 falls associated with voiced codas within these stimuli likely contributed to the longer perceived duration of vowels from this environment; manipulating f0 contours eliminated effects of the original coda on perceived vowel duration. The effects of the production environment on perceived vowel duration suggest a possible perceptual pathway for the voicing effect on vowel duration.

Glottalisation of word-final stops in Australian English unstressed syllables

Glottalisation functions as a cue to coda stop voicelessness in many varieties of English, occurring most commonly for alveolar stops, although varieties differ according to the context and frequency with which glottalisation is used. In Australian English, younger speakers glottalise voiceless coda stops at much higher rates than older speakers suggesting a recent change to the variety, yet this change has only been examined in stressed syllables for stops with alveolar place of articulation. In addition, research has found that glottalisation occurs in a trading relationship with preceding vowel duration to cue coda stop voicing: younger speakers make less use of vowel duration and more use of glottalisation. This study investigates glottalisation as a cue to coda voicing in unstressed syllables, an environment in which coda voicing-related vowel durational differences are already reduced. We examine this phenomenon in two separate datasets of Australian English with reference to stops at three places of articulation to explore dialect-specific distributional patterns and to track the potential progression of change. The results suggest that glottalisation occurs in conjunction with voiceless stops at all places of articulation in the unstressed Australian English contexts examined here. The results also confirm that younger speakers employ glottalisation more than older speakers, and show that females glottalise more than males, both results supporting previous suggestions of a recent change to the variety.

Online perception of glottalized coda stops in American English

In American English, voiceless codas /t/ and /p/ are often glottalized: They have glottal constriction that results in creaky voice on the preceding vowel. Previous claims suggest that such glottalization can serve to enhance /t/ or, more generally, voicelessness of coda stops. In this study, we examine the timecourse of word recognition to test whether glottalization facilitates the perception of words ending in voiceless /t/ and /p/, which is expected if glottalization is in fact enhancing. Sixty American English listeners participated in an eye-tracking study, where they heard resynthesized glottalized and non-glottalized versions of CVC English words ending in /p, t, b, d/ while looking at a display with two words presented orthographically. Target words were presented with a minimal pair differing in place of articulation (e.g., cop-cot), or voicing, (e.g., bat-bad, cap-cab). Although there is little evidence that glottalization facilitates recognition of words ending in /t/ or /p/, there is a strong inhibitory effect: Words ending in voiced stops are recognized more slowly and poorly when the preceding vowel was glottalized. These findings lend little support to a listener-driven, enhancement-based explanation for the occurrence of coda glottalization in American English. On the other hand, they suggest that glottalized instances of coda /t/ and /p/, but not of coda /d/ and /b/, are perceived as equally good variants of these sounds.

Glottalisation as a cue to coda consonant voicing in Australian English

Recent research has suggested that some long vowels exhibit a constraint on voiceless coda induced shortening in Australian English. This constraint has the potential to compromise the vowel length cue to coda voicing, raising questions about how the coda stop voicing contrast is preserved if vowel duration cues are weakened. One possibility is that glottalisation and vowel duration are exploited in a trading relationship as cues to coda voicelessness. We conducted an apparent time study using a highly constrained dataset to determine whether vowel duration and the presence of glottalisation covary as cues to coda stop voicing and whether there is evidence of recent change to the weighting of these cues in Australian English. The incidence of glottalisation and temporal aspects of voiced and voiceless rhymes were analysed in the speech of male and female speakers from two different age groups. Results showed that younger speakers exhibited greater use of glottalisation in voiceless contexts in association with reduced coda voicing-related vowel duration cues compared to older speakers. These finding raise questions about the weighting of glottalisation and vowel duration as cues to coda stop voicing and may suggest change in progress regarding the management of voicing cues in the rhyme.

Online perception of coda glottalization in American English

In American English, voiceless codas /t, p/ are often realized with glottalization on the preceding vowel. Previous claims suggest that such glottalization can serve to enhance /t/ or, more generally, voicelessness of coda stops. This study examines the timecourse of word recognition to test these claims. 40 American English listeners participated in an eye-tracking study, where they heard synthesized glottalized and non-glottalized versions of CVC English words ending in /p, t, b, d/ while looking at a display with two words presented orthographically. Target words were presented with a minimal pair differing in place of articulation (e.g., cop-cot), or voicing (e.g., bat-bad, cap-cab). Our results indicate that listeners fixated to target words ending in /t/ marginally faster when they heard the glottalized version and when the competitor presented was a word ending in /p/. Glottalization did not result in faster fixation to targets for words ending in /p/. We also found a strong inhibitory effect—lower...

Differences in coda voicing trigger changes in gestural timing: A test case from the American English diphthong /aɪ/

We investigate the hypothesis that duration and spectral differences in vowels before voiceless versus voiced codas originate from a single source, namely the reorganization of articulatory gestures relative to one another in time. As a test case, we examine the American English diphthong /aɪ/, in which the acoustic manifestations of the nucleus /a/ and offglide /ɪ/ gestures are relatively easy to identify, and we use the ratio of nucleus-to-offglide duration as an index of the temporal distance between these gestures. Experiment 1 demonstrates that, in production, the ratio is smaller before voiceless codas than before voiced codas; this effect is consistent across speakers as well as changes in speech rate and phrasal position. Experiment 2 demonstrates that, in perception, diphthongs with contextually incongruent ratios delay listeners’ identification of target words containing voiceless codas, even when the other durational and spectral correlates of voicing remain intact. This, we argue, is evidence that listeners are sensitive to the gestural origins of voicing differences. Both sets of results support the idea that the voicing contrast triggers changes in timing: gestures are close to one another in time before voiceless codas, but separated from one another before voiced codas.

Effects of following onsets on voice onset time in English

Voice Onset Time (VOT) in English voiceless stops has been shown to be sensitive to place of articulation (Fischer-Jorgensen 1954), to contextual factors such as the height, tenseness, and duration of the following vowel and the voicing of coda consonants (Klatt 1975, Port & Rotunno 1979), to prosodic factors like stress and pitch (Lisker & Abramson 1967), and also to F0 (McCrea & Morris 2005) and speaking rate (Kessinger & Blumstein 1997, Allen 2003). We report two additional factors involving following consonants. We analyzed 120 /p/- and /k/-initial words produced by 148 Canadian English speakers (n = 17742). VOTs of the initial stops were measured semi-automatically and all other segment durations were measured using forced alignment. The results of a mixed-effects regression support earlier findings that VOT is longer in /k/, directly related to following vowel duration, inversely related to speech rate, longer before tense vowels, and shorter before voiceless codas. Additionally, we find that VOT is shorter when the next syllable starts with a phonetically voiceless plosive (i.e., excluding flapped /t/), and that the most relevant measure of vowel duration includes the duration of postvocalic liquids, even those that are typically analyzed as onsets.

Nuclear pitch accent in American English dialects: Truncation and compression

Dialects of the same language can differ in their use of prosody. We explored realization of nuclear pitch accent in female speech in three regional varieties of American English: Ohio, Wisconsin, and North Carolina. We found that both OH and WI speakers truncate the F0 contours while NC speakers compress them in identical environments. Our second question pertained to the effects of obstruent voicing in syllable coda on the F0 of the preceding nuclear vowel. We admitted three possibilities for the effects of a voiceless coda: shortened vowel duration may (1) “clip” the F0 contour and reduce the dialectal differences; (2) maintain the dialectal differences by preserving the F0 contour shapes found before a voiced coda; or (3) some dialects may “clip” and some may preserve the F0 contour. The results supported the third option. Ohio and Wisconsin speakers “clipped” the F0 contour so that F0 terminal values before a voiceless coda were higher than before a voiced coda. However, North Carolina speakers not only preserved the contours but their F0 terminal values before a voiceless coda were substantially lower than before a voiced coda. The effects of coda voicing on F0 fall were found to augment the dialectal differences.

Effects of speaking rate, sentential position, and coda voicing on formant frequency

This study examines the effects of speaking rate, sentential position, and coda voicing on formant frequency values in English. Some previous studies have found gestural undershoot for formant target values in words with shorter durations, e.g., Lindblom (1963), although other studies have shown little to no effect of duration, e.g., Gay (1978), and the effects of the other factors are less-studied. This study examines formant frequencies of three different vowels (/i/, /e/, and /ae/) in CVC words containing both voiced and voiceless codas produced in three different sentence positions (initial, medial, and final) and three different speaking rates (slow, habitual, and fast). In total, seven speakers (five female and two male) of the Northern dialect of American English produced 1295 tokens, and vowel formant measurements were extracted at 1/3 of the duration of each vowel token. Separate linear regression analyses of the three vowels for the male and female speakers show that F1 and F2 target values do n...

Effects of syllable position on sound change: An aerodynamic study of final fricative weakening

This study explores the differences between the aerodynamic characteristics of onset and coda fricatives. Such differences may account for the weakening of syllable-final fricatives found synchronically and diachronically in a variety of languages. Simultaneous oral pressure, airflow, and audio-signal were obtained for two English speakers. Aerodynamic and acoustic analyses showed that, in comparison with onset fricatives, coda fricatives exhibit (i) a slower oral pressure build-up, (ii) a lower pressure peak, (iii) a delayed onset of audible frication, and (iv) lower intensity of frication. In order to provide high flow for frication, voiceless coda fricatives show earlier glottal opening (i.e., a larger increase in flow at the VC transition) than onset fricatives, suggesting compensatory strategies. The findings suggest that modification of the aerodynamic conditions – velocity and duration of flow through the oral constriction – due to a decreased oral gesture syllable-finally and/or lowered subglottal pressure utterance-finally (and possibly coarticulation with conflicting segments) make audible friction more difficult for speakers to produce and for listeners to detect, leading to final fricative weakening or loss.

Acoustic cues to stop coda voicing contrasts in 1 to 2-year-olds’ American English.

One limitation in understanding the mechanisms underlying children’s variable productions has been the reliance on perception-based segmental transcription. This raises the possibility that children may make covert, acoustic distinctions that are not perceived by the transcriber. Shattuck-Hufnagel et al. [S. Shattuck-Hufnagel, K. Demuth, H. Hanson & S. Stevens, submitted] investigated this issue employing Stevens’ [K. N. Stevens, J. Acoust. Soc. Am. 111, 1872–1891 (2002)] feature-based acoustic model of perception/production. They found that 2,6–3,6-year-olds exhibited systematic acoustic cues to coda voicing contrasts (e.g., dog versus duck): voice-bar was more likely to precede voiced codas, whereas vowel glottalization was more likely to precede voiceless codas. The present study extended this investigation to velar codas in three 1,6–2,6-year-olds and their mothers, showing similar results: both exhibited more use of voice-bar before voiced compared to voiceless codas. For mothers, the duration of voice-bar was also longer for voiced codas, and children showed a trend in this direction. However, only mothers showed a significantly higher use of vowel glottalization before voiceless codas. Thus, although younger children also produce some acoustic cues to voicing distinctions, others take time to become adultlike. Physiological and environmental factors are discussed to explain these results. [Work supported by R01 HD057606.]

Acoustic cues to the voicing contrast in developing coda stop consonants: Toward comparison with adult caregivers.

Hanson and Shattuck‐Hufnagel (Proceedings of Acoustics ‘08) analyzed coda stop consonants for three children from the Imbrie corpus (A. Imbrie, Ph.D. thesis, MIT, 2005). Following the closure, voice bars were frequently observed in [+voice] stops, while [−voice] codas often had a period of noise overlapping the preceding vowel and extending into the closure. Possible explanations for this noise include motoric immaturity, imitation of something in their caregivers’ productions, or the child’s decision to provide additional cues to the voicing contrast (enhancement). In this study, the second possibility was investigated in recordings of each child’s primary caregiver. Coda stop consonants were analyzed in the words “bug,” “tub,” “cup,” and “duck,” specifically the durations of voice bars and vowel‐final noise. As expected, voice bars mainly occur in voiced codas; in contrast, vowel‐final noise occurs mainly with voiceless codas, and its duration and degree of overlap with the vowel/closure vary across subjects. Similar measures for the children will be compared with those of their primary caregivers. [Work supported by NIH Grants Nos. DC00075 and DC 008780, and a NSF CPATH grant.]

Acoustic cues to the voicing contrast in coda stops in the speech of 2-year-olds learning American English

Developing a model of how children acquire the phonological feature contrasts and phonetic cues of their adult speech community requires a detailed understanding of how production changes during development. Earlier studies based on listening provided an initial estimate of these patterns; recent instrumental studies of the acoustic and articulatory details of production during development have revealed some surprises, including covert contrasts between distinctive feature categories (e.g., Scobbie et al., 1996) and of incomplete acquisition of adult feature-cue patterns (Imbrie, 2005), which can be difficult for adults to hear. Few instrumental studies, however, have focused on the child's cue patterns for coda consonants (although see Song and Demuth, in press). Our study of cues to the feature [voice] for coda stops used quantitative acoustic analyses of tokens from the Imbrie corpus of 2-year-old speech to compare child productions with those of their adult caretakers. Results show that a) many children produce a noisy region at the end of the vowel for voiceless codas (in duck, cup) but a long strong voice bar during closure for voiced codas (in bug, tub), and b) these cues may be exaggerated versions of the feature cues of their adult caretakers.

Stability of temporal contrasts across speaking styles in English and Croatian

In this paper, we investigate the effect of clear speech, a distinct, listener-oriented, intelligibility-enhancing mode of speech production, on vowel and stop consonant contrasts along the temporal dimension in English and Croatian. Our previous work has shown that, in addition to enhancing the overall acoustic salience of the speech signal through a decrease in speaking rate and expansion of pitch range, clear speech modifications increased the spectral distances between vowel categories in both languages despite the different sizes of their vowel inventories (+10 in English, five in Croatian). Here, we examine how clear speech affects the duration of English tense (long) vs. lax (short) vowels, English vowels preceding voiced (long) vs. voiceless (short) coda stops, Croatian long vs. short vowels and Croatian and English voice onset time (VOT) duration for voiced and voiceless stops. Overall, the results showed that the proportional distance between the ‘short’ and ‘long’ vowel categories and between the voiced and voiceless stop categories was remarkably stable across the two speaking styles in both languages. These results suggest that, in combination with the spectral enhancement of vowel contrasts, language-specific pronunciation norms along the temporal dimension are maintained in clear and conversational speech.