Consonant Voicing Landmark Detection

The landmark detection is the first step of a knowledge-based speech recognition system. For landmark detection, it is necessary to extract some useful landmark acoustic parameters. In this project, we do a extensive study about consonant voicing landmark detection.

Consonant voicing and devoicing

In this project, we investigate consonant voicing and devoicing problems and claim that several acoustic parameters are effective.

Consonant Voicing Investigation

Obstruct consonant could be detected by energy difference profiles. In thisproject, we investigate algorithm to detect three types of landmarks, [stopclosure], [stop release], [fricative closure]. As a result, Overall detection ratesfor stop closure and release are 71.3% and 82.3% for obstruent landmarks inTIMIT, and fricatives yield 83.2% and 81.9% respectively.

Incomplete neutralization in African American English: The case of final consonant voicing

In many varieties of African American English (AAE), glottal stop replacement and deletion of word-final /t/ and /d/ results in consonant neutralization, while the underlying voicing distinction may be maintained by other cues, such as vowel duration. Here, I examine the relationship between vowel duration, final glottal stop replacement, and deletion of word-final /t, d/ to determine whether the phonological contrast of consonant voicing is maintained through duration of the preceding vowel. Data come from conversational interviews of AAE speakers in North Carolina, Tennessee, and Washington, DC. Results indicate that glottalization and deletion of word-final /t/ and /d/ are widespread across the speakers in the analysis. Additionally, the duration of vowels is significantly longer before underlying /d/ than /t/ for consonant neutralized contexts, thus showing that duration, normally a secondary cue to final voicing, may be becoming a primary cue in AAE.