PubMed Health⌕ Search
Home/Archive/PubMed

PubMed · 10530027

A model of auditory perception as front end for automatic speech recognition.

J Tchorz·B Kollmeier
View original source ↗Export metadata ↓

Abstract

A front end for automatic speech recognizers is proposed and evaluated which is based on a quantitative model of the "effective" peripheral auditory processing. The model simulates both spectral and temporal properties of sound processing in the auditory system which were found in psychoacoustical and physiological experiments. The robustness of the auditory-based representation of speech was evaluated in speaker-independent, isolated word recognition experiments in different types of additive noise. The results show a higher robustness of the auditory front end in noise, compared to common mel-scale cepstral feature extraction. In a second set of experiments, different processing stages of the auditory front end were modified to study their contribution to robust speech signal representation in detail. The adaptive compression stage which enhances temporal changes of the input signal appeared to be the most important processing stage towards robust speech representation in noise. Low-pass filtering of the fast fluctuating envelope in each frequency band further reduces the influence of noise in the auditory-based representation of speech.

Explore related subjects

Auditory PerceptionAutomatismHumansModels, BiologicalModels, PsychologicalNoiseSpeech PerceptionTime FactorsJournal ArticleResearch Support, Non-U.S. Gov't

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

J Tchorz, B Kollmeier. 1999. A model of auditory perception as front end for automatic speech recognition.. https://doi.org/10.1121/1.427950

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related citations

PubMed2007

Hearing lips and seeing voices: how cortical areas supporting speech production mediate audiovisual speech perception.

Observing a speaker's mouth profoundly influences speech perception. For example, listeners perceive an "illusory" "ta" when the video of a face producing /ka/ is dubbed onto an audio /pa/. Here, we show how cortical areas supporting speech production mediate this illusory percept and audiovisual (AV) speech perception more generally. Specifically, cortical activity during AV speech perception occurs in many of the same areas that are active during speech production. We find that different perceptions of the same syllable and the perception of different syllables are associated with different distributions of activity in frontal motor areas involved in speech production. Activity patterns in these frontal motor areas resulting from the illusory "ta" percept are more similar to the activity patterns evoked by AV(/ta/) than they are to patterns evoked by AV(/pa/) or AV(/ka/). In contrast to the activity in frontal motor areas, stimulus-evoked activity for the illusory "ta" in auditory and somatosensory areas and visual areas initially resembles activity evoked by AV(/pa/) and AV(/ka/), respectively. Ultimately, though, activity in these regions comes to resemble activity evoked by AV(/ta/). Together, these results suggest that AV speech elicits in the listener a motor plan for the production of the phoneme that the speaker might have been attempting to produce, and that feedback in the form of efference copy from the motor system ultimately influences the phonetic interpretation.

Auditory Perception↗
PubMed2006

Top-down processes during auditory phoneme categorization in dyslexia: a PET study.

While persistence of subtle phonological deficits in dyslexic adults is well documented, deficit of categorical perception of phonemes has received little attention so far. We studied learning of phoneme categorization during an activation H(2)O(15) PET experiment in 14 dyslexic adults and 16 normal readers with similar age, handedness and performance IQ. Dyslexic subjects exhibited typical, marked impairments in reading and phoneme awareness tasks. During the PET experiment, subjects performed a discrimination task involving sine wave analogues of speech first presented as pairs of electronic sounds and, after debriefing, as syllables /ba/ and /da/. Discrimination performance and brain activation were compared between the acoustic mode and the speech mode of the task which involved physically identical stimuli; signal changes in the speech mode relative to the acoustic mode revealed the neural counterparts of phonological top-down processes that are engaged after debriefing. Although dyslexic subjects showed good abilities to learn discriminating speech sounds, their performance remained lower than those of normal readers on the discrimination task over the whole experiment. Activation observed in the speech mode in normal readers showed a strongly left-lateralized pattern involving the superior temporal, inferior parietal and inferior lateral frontal cortex. Frontal and parietal subparts of these left-sided regions were significantly more activated in the control group than in the dyslexic group. Activations in the right frontal cortex were larger in the dyslexic group than in the control group for both speech and acoustic modes relative to rest. Dyslexic subjects showed an unexpected large deactivation in the medial occipital cortex for the acoustic mode that may reflect increased effortful attention to auditory stimuli.

Auditory Perception↗