Vowel Recognition

セブアノ語は固有語では 3 母音体系を有し、フィリピンでタガログ語に次いで母語話者
の多い言語である。本稿ではセブアノ語を母語とする日本語学習者を対象に、日本語の母
音の聞き誤りについて予備的な考察を行った。
特殊拍の誤りが最も多く、次いで母音の誤りが多かった。特殊拍の誤りは母語を問わず
見られるものであることから、母音の誤りは本稿で扱った学習者に特徴的なものであると
言えるだろう。母音の聞き誤りのほとんどはイ段・エ段間、ウ段・オ段間に見られたが、
その誤りの傾向には個人差が見られた。本稿では 3 名の学習者について考察を行ったが、1
名はウ段・オ段間、特に拗音をともなう音で誤りが多かった。他の 1 名は半狭母音を聞い
た際に狭母音と判断する誤りが多かった。もう 1 名は母音の聞き誤りが少なかった。
フィリピンは多言語国家であるが、フィリピン諸語で日本語よりも母音が少ない言語は
ほかにも多く存在する。セブアノ語以外を母語とする学習者や、母音の発話の面からの検
証も必要であると考える。

It can be assumed that native Cebuano speakers who are learning the Japanese language have
trouble in differentiating the different Japanese vowels. This is because the Cebuano language has
only three vowels: /i/, /a/, and /u/.
This study aims to observe the method by which Cebuano native speakers recognize Japanese
language vowels.
I recorded and analyzed the data of three learners through the word quizzes on which they took
dictations. The result shows that some of them find it difficult to differentiate /o/ from /u/ and /e/
from /i/; however, this depends on the abilities of the individuals.

—This paper presents a signal processing approach to analyse and identify accent discriminative features of four groups of English as a second language (ESL) speakers, including Chinese, Indian, Japanese, and Korean. The features used for speech recognition include pitch, stress, formant frequencies, the Mel frequency coefficient, log frequency coefficient, and the intensity and duration of vowels spoken. This paper presents our study using the Matlab Speech Analysis Toolbox, and highlights how data processing can be automated and results visualised. The proposed algorithm achieved an average success rate of 57.3% in identifying vowels spoken in a speech by the four non-native English speaker groups.

RESUME __________________________________________________________________________________________________________

Nous présentons une évaluation perceptive des voyelles françaises /y/ et /u/ produites
par des apprenants hispanophones et japonophones dans le cadre du projet
InterPhonologie du Français Contemporain (IPFC). Afin d’évaluer la qualité de réalisation
de leurs productions, nous avons effectué deux expériences dans lesquelles 58 natifs ont
dû identifier la voyelle (/y-u/) de monosyllabes produits par un groupe de natifs et deux
groupes d’hispanophones (Expérience 1) et un groupe de natifs et deux groupes de
japonophones (Expérience 2) en répétition et en lecture. Les résultats montrent
globalement un effet de population – meilleure identification pour les natifs, – de tâche –
meilleure identification en répétition – et, pour les hispanophones, de voyelle –
meilleure identification pour /u/ que pour /y/. Ils révèlent également des différences
entre hispanophones et japonophones, ainsi que, pour chaque population, entre les deux
sous-groupes d’apprenants, qui se distinguent par le degré d’exposition au français.

ABSTRACT _________________________________________________________________________________________________________

The /y-u/ vowels in the IPFC project: a perceptual assessment of native, Spanish
and Japanese learners’ productions.
We present a perceptual study of the French vowels /y/ and /u/ produced by Spanish
and Japanese learners, in the framework of the InterPhonology of Contemporary French
(IPFC) project. To evaluate the quality of realization of their productions, we carried out
two experiments in which 58 native listeners had to identify the vowel (/y-u/) of
monosyllables produced by one group of native speakers, two categories of Spanish
learners, and two categories of Japanese learners in a repetition and a reading task. The
results globally show a population effect – a better identification for the natives – a task
effect – a better identification in the repetition task – and, for the Spanish learners, a
vowel effect – a higher identification rate for /u/ than for /y/. They also reveal
differences between Spanish and Japanese learners, and between the two categories of
learners within each group, which differ in terms of degree of L2 exposure.

MOTS-CLES : français langue étrangère, apprenants espagnols, apprenants japonophones,
évaluation perceptive, voyelles arrondies, phonologie L2.

KEYWORDS : L2 French, Spanish learners, Japanese learners, perceptual assessment,
rounded vowels, L2 phonology.

Automatic speech recognition (ASR) has made great strides with the development of digital signal processing hardware and software especially using English as the language of choice. In this paper, a new feature extraction method is presented to identify vowels recorded from 80 Malaysian speakers. The features were obtained from vocal tract model based on bandwidth (BW) approach. Bandwidth approach identifies frequency bands based on the first peak of vowel frequency responses. Mean and maximum energies were calculated from these Bandwidth frequency bands. Classification results from bandwidth approach were compared with the first 3-formant features using Linear Predictive method. A multi-layer perceptron (MLP) and multinomial logistic regression (MLR) were used to classify the vowels. MLR and MLP shows comparable classification results for BW approach of 96.40% and 96.59% respectively. Bandwidth approach obtained 5.49% higher classification rate than 3-formant features using MLP.