単語を認識単位とした日本語の大語彙連続音声認識  [in Japanese] Word - based approach to large - vocabulary continuous speech recognition for Japanese  [in Japanese]

Access this Article

Search this Article

Author(s)

Abstract

我々は先に,日本人が感覚的に捉えている単語単位を,既存の形態素解析プログラムの出力である形態素単位との統計的対応関係から自動推定する方法を提案し,それを認識および発声の単位とする離散単語発声の日本語ディクテーションシステムを構築した.今回,この人間の考える単語単位を連続音声認識の認識単位としても利用することを試み、特に,他の大語紮連続音声認識システムで用いられる事の多い形態素単位と比較し,その有効性について調査した.また,認識単位の定義が一意に決まらない現状を踏まえて,日本語の連続音声認識システムの評価方法を提案するとともに,ついて報告する.男女各10名に対する認識実験の結果,文字誤り率 3%,単語誤り率 4.3%が得られた.さらに,句読点の自動挿入方法や,未知語モデルを使った単語 N?gramによる単語単位の自動分割方法などについても述べる.In This paper, we discuss a word-based Japanese continuous dictation system. We have previously proposed a statistical method for segmenting a text into words on the basis of human intuition, and developed an isolated-word-based Japanese dictation system. By comparing this word unit used for the isolated word recognition with grammatical units, we show that this unit is also very useful for continuous speech recognition. Evaluation of the performance of this continuous dictation system showed that the character error rate was 3%, and that the word error rate was 43%. We also present a method for inserting punctuation marks in spoken texts automatically, and a method for segmenting Japans text into words by using an N-gram model, focusing on how to handle unknown words.

In this paper, we discuss a word-based Japanese continuous dictation system. We have previously proposed a statistical method for segmenting a text into words on the basis of human intuition, and developed an isolated-word-based Japanese dictation system. By comparing this word unit used for the isolated word recognition with grammatical units, we show that this unit is also very useful for continuous speech recognition. Evaluation of the performance of this continuous dictation system showed that the character error rate was 3%, and that the word error rate was 4.3%. We also present a method for inserting punctuation marks in spoken texts automatically, and a method for segmenting Japanese text into words by using an N-gram model, focusing on how to handle unknown words.

Journal

  • IPSJ SIG Notes

    IPSJ SIG Notes 1998(12(1997-SLP-020)), 17-24, 1998-02-05

    Information Processing Society of Japan (IPSJ)

References:  8

Cited by:  15

Codes

  • NII Article ID (NAID)
    110002917023
  • NII NACSIS-CAT ID (NCID)
    AN10442647
  • Text Lang
    JPN
  • Article Type
    Journal Article
  • ISSN
    09196072
  • Data Source
    CJP  CJPref  NII-ELS  IPSJ 
Page Top