JPH024920B2 - - Google Patents

Description

Translated fromJapanese

【発明の詳細な説明】産業上の利用分野 本発明は音声認識装置に関するものである。[Detailed description of the invention]Industrial applications The present invention relates to a speech recognition device.

従来例の構成とその問題点 音声認識装置は、人間の音声命令によつて対象
の機器の動作を制御することを目的とする命令入
力装置の一種であり、○…覗犧郛紊侶盈Configuration of conventional examples and their problems A voice recognition device is a type of command input device whose purpose is to control the operation of target equipment using human voice commands.

Claims (1)

Translated fromJapanese

【特許請求の範囲】１ 入力音声の音韻を任意の短時間区間毎に識別
して音韻系列を出力する音韻識別手段と、認識し
ようとする音声単位が有すると予想される標準的
な音韻系列が予め登録された音声単位標準パター
ンと、各音韻間の類似性を表わす音韻類似度行列
と、前記音韻識別手段から出力される音韻系列と
前記音声単位標準パターンと前記音韻類似度行列
とを用いて入力音声の音韻性と前記音声単位標準
パターンの音韻性との間の類似度を前記短時間区
間の音韻毎に表現する音声単位音韻比較行列と、
この音声単位音韻比較行列によつて表わされると
ころの音韻類似性を音声単位毎に求める音声単位
類似度計算手段とを備え、予め音韻別に雑音の強
度・特性に応じた複数組の重み付け係数を用意し
ておき、入力音声に重畳する雑音の強度及び特性
を検出する雑音検出手段から得られる情報に基づ
いて、適宜特定の重み付け係数の組を選択し、前
記音声単位類似度計算手段により前記重み付け係
数を用いて音声単位毎に音韻類似度を計算する構
成とした音声認識装置。２ 入力音声を任意の短時間区間毎に音響分析し
てその音響特徴系列を出力する音響分析手段と、
認識しようとする音声単位が有すると予想される
標準的な音響特徴系列が予め登録された音声単位
標準パターンと、前記音響分析手段から出力され
る音響特徴系列と前記音声単位標準パターンとを
用いて入力音声と音声単位標準パターンとの間の
類似度を前記短時間区間の入力音声毎に表現する
音声単位音響比較行列と、この音声単位音響比較
行列によつて表わされるところの音響類似度を音
声単位毎に求める音声単位類似度計算手段とを備
え、予め音響特徴量別に雑音の強度・特性に応じ
た複数組の重み付け係数を用意しておき、入力音
声に重畳する雑音の強度及び特性を検出する雑音
検出手段から得られる情報に基づいて、適宜特定
の重み付け係数の組を選択し、前記音声単位類似
度計算手段により前記重み付け係数を用いて音声
単位毎に音響類似度を計算する構成とした音声認
識装置。[Claims] 1. A phoneme identification means for identifying the phonemes of an input speech every arbitrary short time interval and outputting a phoneme sequence, and a standard phoneme sequence expected to be possessed by a phonetic unit to be recognized. Using a pre-registered phoneme unit standard pattern, a phoneme similarity matrix representing the similarity between each phoneme, a phoneme sequence output from the phoneme identification means, the phoneme standard pattern, and the phoneme similarity matrix, a speech unit phoneme comparison matrix that expresses the degree of similarity between the phonology of the input speech and the phonology of the speech unit standard pattern for each phoneme of the short time interval;
A speech unit similarity calculating means is provided for calculating the phoneme similarity represented by this speech unit phoneme comparison matrix for each speech unit, and multiple sets of weighting coefficients are prepared in advance for each phoneme according to noise intensity and characteristics. Based on the information obtained from the noise detection means that detects the intensity and characteristics of the noise superimposed on the input speech, a specific set of weighting coefficients is appropriately selected, and the weighting coefficients are determined by the speech unit similarity calculation means. A speech recognition device configured to calculate phonological similarity for each speech unit using . 2. acoustic analysis means for acoustically analyzing input speech for each arbitrary short time interval and outputting the acoustic feature sequence;
A speech unit standard pattern in which a standard acoustic feature sequence that is expected to be possessed by the speech unit to be recognized is registered in advance, an acoustic feature series output from the acoustic analysis means, and the speech unit standard pattern are used. A speech unit acoustic comparison matrix that expresses the similarity between the input speech and the speech unit standard pattern for each input speech in the short time period, and a speech unit acoustic comparison matrix that expresses the acoustic similarity expressed by this speech unit acoustic comparison matrix. It is equipped with a speech unit similarity calculation means that is calculated for each unit, prepares in advance multiple sets of weighting coefficients according to the intensity and characteristics of noise for each acoustic feature value, and detects the intensity and characteristics of noise superimposed on input speech. Based on the information obtained from the noise detection means, a specific set of weighting coefficients is selected as appropriate, and the acoustic similarity calculation means calculates acoustic similarity for each speech unit using the weighting coefficients. Speech recognition device.