Connine, “ Missing information in spoken word recognition: Nonreleased stop consonants,” J. In particular, we provide the detailed analysis of incomplete stop consonants 4 4. with a specific focus on the robustness of the stop detectors in the presence of incomplete stops. Garofolo, “ TIMIT acoustic-phonetic continuous speech corpus,” LDC, Philadelphia (1993). In this work, we investigate the complete and incomplete stops in the TIMIT database 10 10. complete stops accounted for only about 45% of the identified stops. House, “ The duration of American-English stop consonants: An overview,” J. For example, in a comprehensive study by Crystal and House, 3 3. But corpus studies (in English) have shown that acoustic implementation of complete stops is only a fraction of the possibilities. which is defined as one that should include an identifiable closure portion followed by a burst release. However, all these approaches for stop detectors implicitly assume that a stop should be a complete stop, 3 3. Niyogi, “ Modeling the temporal dynamics of distinctive feature landmark detector for speech recognition,” J. These features are in turn being used within novel automatic speech recognition frameworks such as those based on landmark detectors. have been used to capture a period of extremely low energy (corresponding to the period of closure) followed by a sharp, broadband signal (corresponding to the release). Montresor, “ Detection of stop consonants with the wavelet transform,” in Proceedings of IEEE-SP International Symposium on Time-Frequency and Time-Scale Analysis (1994), pp. Sondhi, “ Detecting stop consonants in continuous speech,” J. Raphael, “ Stop-consonant recognition: Release bursts and formant transitions as functionally equivalent, context-dependent cues,” Percept. Mueller, “ Acoustic-phonetic features for the automatic classification of stop consonants,” IEEE Trans. For example, to detect stop consonants, spectral and temporal features, 5,6 5. To address this problem of ASR, researchers have proposed several alternative features and models to detect stop consonants. These short-dynamic acoustic variations make the state-of-the-art hidden Markov model based automatic speech recognizer (ASR) incapable of performing accurate fine phoneme distinctions for this class of sounds. which sometimes results in the absence of a clear stop release or burst. Beckman ( Cambridge University Press, Cambridge, 1990), pp. Goldstein, Tiers in Articulatory Phonology, With Some Implications for Casual Speech, Papers in Laboratory Phonology, edited by J. However, in conversational or even read speech, this acoustic signature of a stop is not always apparent due to intergestural overlap, 2 2. Coleman, Acoustics of American English Speech: A Dynamic Approach ( Springer-Verlag, Berlin, 1993), pp. A stop consonant (/b/, /d/, /g/ and /p/, /t/, /k/ ) is produced when there is complete closure of the articulators, stopping the airflow in the vocal tract, followed by a release or burst of air. The implications of these data for the description of voicing errors also are discussed.Stop consonants in English speech have been a topic of research over the past few decades, particularly to better understand and analyze the dynamic and highly speaker- and context-dependent nature of these sounds. These various results are interpreted to mean that some phonologically disordered children may have less mature speech timing control. Perceptual analysis indicated significantly more voicing errors in the initial final stops of phonologically disordered children. However, like normal subjects, they demonstrated differential vowel and consonant closure durations in voiced and voiceless contexts. In the word-final voicing contrast, the phonologically disordered children evidenced longer consonant closure durations and less voicing during consonant closure than did normal subjects. Results indicated that some phonologically disordered subjects failed to differentiate VOT in word-initial voiced and voiceless stops, whereas others produced much longer VOTs for voiceless stops than did control subjects. In addition, listener transcriptions were employed for perceptual analysis. Measurements of voice onset time, vowel duration, consonant closure duration, and voicing during consonant closure were made from spectrograms. Speech timing of nine phonologically disordered and nine normally developing children was investigated for the voicing contrasts of word-initial and word-final stop consonants.

Onset stop closure durations