The heart of a speech recognition system is the acoustic model of sub-word units (e.g., phonemes). In this work we discuss refinements of the stochastic segment model, an alternative to hidden Markov models for representation of the acoustic variability of phonemes. We concentrate on mechanisms for better modelling time correlation of features across an entire segment. Results are presented for speaker-independent phoneme classification in continuous speech based on the TIMIT database.

HEATH, HANS-MARTIN DEDERDING Explanation in Linguistics The Logical Problem of Language Acquisition … traditional answer to this question is that phonemes are the ultimate units of phonol¬ ogical structure … words. Now, if we take the view that phonemes are the ultimate indivisible units of phonological

recognized by a hearer, if interchanged) are the phonemes of the language. The phonemes in turn are the constituents ... information of interest is bundled together in a single, extended text segment (as in the case of earthquake ... sentiment) of a discourse segment , then symbolic representations need not play any role in the computational ...

contain phonemes (or the spatial–gestural equivalent in sign languages), and all spoken languages include both consonant and vowel phonemes. Phonemes are studied A phoneme () is a set of similar speech sounds that are perceptually regarded by the speakers of a language as a single basic sound—the smallest possible phonetic unit—that helps distinguish one word from another. All languages contain phonemes (or the spatial–gestural equivalent in sign languages), and all spoken languages include both consonant and vowel phonemes. Phonemes are studied under phon Phonemes that are contrastive in certain environments may not be contrastive in all environments. In the environments where they do not contrast, the contrast is said to be neutralized. In these positions it may become less clear which phoneme a given phone represents. Absolute neutralization is a phenomenon in which a segment of the underlying representation is not realized in any of its phonetic representations (surface forms). The term was introduced by Paul Kiparsky (1968), and contrasts with contextual neutralization where some phonemes are not contrastive in certain environments. Some phonologists prefer not to specify a unique phoneme in such cases, since to do so would mean providing redundant or even arbitrary information—instead they use the technique of underspecification. An archiphoneme is an object sometimes used to represent an underspecified phoneme. An example of neutralization is provided by the Russian vowels /a/ and /o/. These phonemes are contrasting in stressed syllables, but in unstressed syllables the contrast is lost, since both are reduced to the same sound, usually [ə] (for details, see vowel reduction in Russian). In order to assign such an instance of [ə] to one of the phonemes /a/ and /o/, it is necessary to consider morphological factors (such as which of the vowels occurs in other forms of the words, or which inflectional pattern is followed). In some cases even this may not provide an unambiguous answer. A description using the approach of underspecification would not attempt to assign [ə] to a specific phoneme in some or all of these cases, although it might be assigned to an archiphoneme, written something like ⫽A⫽, which reflects the two neutralized phonemes in this position, or {a|o}, reflecting its unmerged values. A somewhat different example is found in English, with the three nasal phonemes /m, n, ŋ/. In word-final position these all contrast, as shown by the minimal triplet sum /sʌm/, sun /sʌn/, sung /sʌŋ/. However, before a stop such as /p, t, k/ (provided there is no morpheme boundary between them), only one of the nasals is possible in any given position: /m/ before /p/, /n/ before /t/ or /d/, and /ŋ/ before /k/, as…

This paper evaluates a direct speech translation Method with waveforms using the Inductive Learning method for short conversation. The method is able to work without conventional speech recognition and speech synthesis because syntactic expressions are not needed for translation in the proposed method. We focus only on acoustic characteristics of speech waveforms of source and target languages without obtaining character strings from utterances. This speech translation method can be utilized for any language because the system has no processing dependent on an individual character of a specific language. Therefore, we can utilize the speech of a handicapped person who is not able to be treated by conventional speech recognition systems, because we do not need to segment the speech into phonemes, syllables, or words to realize speech translation. Our method is realized by learning translation rules that have acoustic correspondence between two languages inductively. In this paper, we deal with a translation between Japanese and English.

derivative follows, metathesis has occurred in the phonemes of the root syllable attended by a contraction of the vowels of root and (derivative) suffix syllables Metathesis ( mə-TATH-ə-siss; from Greek μετάθεσις, from μετατίθημι "to put in a different order") is the transposition of sounds or syllables in a word or of words in a sentence. Most commonly, it refers to the interchange of two or more contiguous segments or syllables, known as adjacent metathesis or local metathesis: anemone > **anenome (onset consonants of adjacent syllables) cavalry > **calv b…

negative of a proposition by pronouncing the phonemes of the affirmative in reverse order, even though it is very easy for us to understand what such a process

Perception of temporal order in vowel sequences with and without formant transitions. Temporal-order perception of phoneme segments in running speech is much superior to temporal-order perception in repeating vowel sequences. The more rapid rates possible in running speech may be due largely to the presence of formant transitions. In a series of five experiments we observed that many temporal-order misjudgements of repeating vowels can be explained in terms of auditory stream segregation, triggered for the most part by discontinuities in first-formant frequencies of adjacent vowels. Streaming, however, can be suppressed by formant transitions appropriate for the perception of stop consonants and by continuous transitions resembling those in coarticulated vowels. At rapid sequence rates, when the constraints of auditory streaming are removed, correct temporal-order identification is limited by linguistic transformations of vowels into other phoneme segments. Published in Journal of experimental psychology. Human perception and performance (1975)

While both segmental and suprasegmental aspects of words have been recognised as potential factors influencing their iconic interpretations, how these components collectively drive the associations of form and affective meaning remains elusive. The current study addressed this issue in a lexical tonal language, Standard Chinese, where suprasegmental pitch information distinguishes word meanings. Specifically, we investigated how phonemic information at both the segmental level (i.e., vowels and consonants) and suprasegmental level (i.e., lexical tones) may influence native Standard Chinese listeners' rating of auditory stimuli's emotional arousal and valence in two-alternative forced-choice tasks. The results indicated a consistent correlation between tones and the perceived arousal and valence ratings of the tone-carrying nonce words. At the segmental level, consonants were more consistently associated with arousal, while vowels correlated with valence. Furthermore, lexical tones were more influential than segmental phonemes in biasing listeners' rating of affective meanings. Regarding arousal ratings, nonce words with falling and rising tones tended to be rated with higher arousal than those with high- and low-dipping tones. Additionally, those with an onset /t/ were rated higher in arousal than those with /n/. Regarding valence ratings, nonce words with falling and low-dipping tones were more likely to receive negative ratings than those with high and rising tones. Moreover, stimuli with /u/ were more inclined to be perceived negatively than those with /i/. Though subtle and sporadic, these findings support the universal tendency of affective iconicity across segments and suprasegmental tones.