Elsevier

Speech Communication

Volume 51, Issue 1, January 2009, Pages 76-82
Speech Communication

Development of familiarity-controlled word lists 2003 (FW03) to assess spoken-word intelligibility in Japanese

https://doi.org/10.1016/j.specom.2008.07.002Get rights and content

Abstract

A new set of “Familiarity-controlled word lists 2003” (FW03) has been developed for a spoken-word intelligibility test in Japanese. FW03 consists of 20 lists of 50 words in four word-familiarity ranks (i.e., 4000 words in total). The entropy of (a) initial moras and (b) sequences consisting of a vowel and a following consonant was maximized in the word lists within each word-familiarity rank. FW03 is now published with speech files of the 4000 words spoken by two male and two female Japanese. The word intelligibility of FW03 was measured with the speech files at various signal-to-noise ratios. In addition to the signal-to-noise ratio effects, strong word-familiarity effects were observed in terms of word intelligibility, indicating that word familiarity is well controlled in FW03. FW03 enables us to measure word intelligibility in several word-familiarity ranks that correspond to the degree of lexical information.

Introduction

Speech intelligibility is assessed using several linguistic units such as phonemes, syllables, words or sentences. Various test materials have been proposed for measuring speech intelligibility with these units. For example, there are monosyllabic word materials such as the phonetically balanced word list (PB) (Egan, 1948), the modified rhyme test (MRT) (House et al., 1965), and the diagnostic rhyme test (DRT) (Voiers, 1983). There are also sentence materials such as the test sentences for speech perception in noise (SPIN) (Bilger et al., 1984, Kalikow et al., 1977), the connected speech test (CST) (Cox et al., 1987), the topic-related everyday sentences developed by The City University of New York (Boothroyd et al., 1988), the hearing in noise test (HINT) (Nilsson et al., 1994), and the semantically unpredictable sentence test (SUS) (Benoit et al., 1996). These test materials relate to the English language.

Fewer speech materials have been proposed for Japanese than for English. However, the Japan Audiological Society proposed syllable lists named 57-S (Japan Audiological Society, 1983) and 67-S (Japan Audiological Society, 1987). 57-S contains five lists of 50 monosyllables and six lists consisting of six digits (2, 3, 4, 5, 6, and 7). 67-S contains eight lists of 20 monosyllables and the same digit lists as 57-S. 67-S is most frequently used to assess hearing abilities and evaluate the fitting of hearing aids.

However, although almost all Japanese monosyllables are meaningful, they are sometimes recognized as nonsense monosyllables especially when they are heard in isolation without a context. This means that an assessment using Japanese monosyllables is not good for controlling lexical information that must be used for speech communication. It is well known that lexical information of a word (e.g., Amano, 1993) as well as contextual information in a sentence (e.g., Connine, 1987) usually improve speech intelligibility and play a major role in speech communication. Therefore, an assessment using words or sentences would be more appropriate than one using monosyllables when measuring the ability of central levels of processing in speech communication. This paper focuses an assessment with words.

In consideration of the appropriateness of an assessment with words, Yonemoto et al. (1989) proposed word lists (TY-89) in Japanese. TY-89 contains 50 two-syllable words and 50 three-syllable words. These words were selected from a list of child vocabulary that everyone should know. TY-89 is used as frequently as 67-S for measuring the ability to hear speech.

However, as Yonemoto (1995) pointed out, TY-89 contains certain words that are hard to recognize. That is, some words always have a lower intelligibility score than others. There are two reasons for this problem. One reason is that word familiarity is not well controlled and it is low for some words in TY-89. Word familiarity is a subjective rating value indicating how familiar a person is with each word. It is typically obtained by averaging participants’ familiarity ratings for a word using a 7-point scale (1: most unfamiliar, 7: most familiar). Controlling word familiarity is very important in test materials, because word recognition depends heavily on word familiarity. That is, the higher the word familiarity is, the more correctly and quickly a spoken word is recognized (Amano and Kondo, 1999, Amano et al., 1999). The other reason is that the phonetic balance is not controlled in TY-89. Since each phoneme differs in terms of its recognizability and a phoneme is a unit constituting a word, the intelligibility score of a word is affected by the phonemes included in the word. Therefore, to assess the speech intelligibility of words with various phonemes, the phonetic balance should be controlled in test materials.

The two reasons given above probably result in an inequality of word items in TY-89. Some words inherently have low intelligibility. To overcome these problems, a new set of word lists (familiarity-controlled word lists 2003, hereafter FW03) was proposed for spoken-word intelligibility tests. Word familiarity was taken into consideration when developing FW03, because it has a strong effect on spoken-word recognition. That is, there is a strong tendency for the recognition accuracy and speed to become higher as the word familiarity increases (Amano and Kondo, 1999). Although word frequency has the same tendency, it has a weaker effect than word familiarity with regard to word recognition (Amano and Kondo, 2000). Therefore, word frequency was not used when developing FW03.

Word familiarity has a very high correlation (r = .958, N = 10,515) when measured over a number of years and in different places (Amano et al., 2007). This means that word familiarity is fairly stable, although it is a subjective measure. This characteristic is another reason why word familiarity was used for developing FW03.

Phonetic balance was also considered when developing FW03, because the intelligibility score of a word must be related to the variety of phonemes included in the word. Previous word lists were also developed considering the phonetic balance. For example, Egan (1948) controlled the distribution of phonemes when developing the PB word list. A mathematical method was used to achieve the phonetic balance in this study.

By controlling these two factors, a reasonable word intelligibility score can be obtained with FW03. This paper describes the procedure for developing FW03 and evaluating word intelligibility with FW03.

Section snippets

Word candidates

Word candidates (13,607) were selected from a word-familiarity database (Amano and Kondo, 1999) with about 88,000 word entries, which were derived from all the word entries in a medium sized Japanese dictionary. The conditions for selection were as follows:

  • 1.

    Word length is usually counted in moras in Japanese (for the definition of mora, see Otake and Cutler, 1996, Vance, 1987). The word length was set at four moras when selecting words because four-mora words are the most frequently occurring

Word intelligibility of FW03

An experiment was conducted to measure the word intelligibility of FW03 at various signal-to-noise ratios. If the development of FW03 is successful, the effect of word familiarity on the word intelligibility of FW03 would be clearly observable as well as the effect of the signal-to-noise ratio.

Conclusion

A set of new word lists, FW03, was developed for a word intelligibility test in Japanese. FW03 consists of 20 lists with 50 words in four word-familiarity ranks. FW03 is a useful set of word lists for the assessment of hearing ability concerning lexical information processing in spoken-word recognition, because the word familiarity and phonetic balance are properly controlled.

References (28)

  • A. Boothroyd et al.

    Voice fundamental frequency as an auditory supplement to the speechreading of sentences

    Ear Hear.

    (1988)
  • R.M. Cox et al.

    Development of the connected speech test (CST)

    Ear Hear.

    (1987)
  • J.P. Egan

    Articulation testing methods

    Laryngoscope

    (1948)
  • A.S. House et al.

    Articulation-testing methods: consonantal differentiation with a closed-response set

    J. Acoust. Soc. Amer.

    (1965)
  • Cited by (45)

    • Brain activations while processing degraded speech in adults with autism spectrum disorder

      2021, Neuropsychologia
      Citation Excerpt :

      In the fMRI and post-test, a full sentence instead of a word was presented on each trial. In the pre-test, we selected 70 four-mora words from the most familiar category in the Japanese familiarity-controlled word list (Amano et al., 2009). Morae, rather than syllables, are the units used in phonology to form the basis of the Japanese sound system.

    • In-home auditory training using audiovisual stimuli on a tablet computer: Feasibility and preliminary results

      2020, Auris Nasus Larynx
      Citation Excerpt :

      Audiovisual recordings of eight lists of 50 four-mora Japanese words (400 words in total) spoken by a trained Japanese female speaker were made. The word lists were selected from phonetically-balanced Japanese word lists developed for a spoken-word intelligibility test (FW03) [15,16] that consist of four word-familiarity ranks (lower, lower-middle, upper-middle, and high familiarities). The recorded word lists were obtained from the Speech Resources Consortium (http://research.nii.ac.jp/src/en/index.html).

    • Effects of inter-word pauses on speech intelligibility under long-path echo conditions

      2018, Applied Acoustics
      Citation Excerpt :

      The results revealed that word intelligibility of high familiarity words is higher than that of low familiarity words, even when the high familiarity words overlap. This tendency is identical to the case when a high familiarity word is presented under low signal-to-noise ratio (SNR) conditions [10] or in a reverberation environment [11]. However, in a real situation, high familiarity words are not always familiar to all listeners, e.g., the name of a place for non-locals.

    View all citing articles on Scopus

    Parts of this research were presented at the 9th Western Pacific Acoustic Conference, Seoul, Korea, June 26–28, 2006.

    View full text