VOICE RECOGNITION DEVICE, VOICE RECOGNITION METHOD AND RECORDING MEDIUM

• Main Author: KANDA NAOYUKI
• Format: Patent
• Language: English; Japanese
• Published: 07.01.2021
• Subjects: ACOUSTICS; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

To precisely perform voice recognition and speaker identification from voices of a plurality of speakers without requiring speaker samples in advance.SOLUTION: The present invention relates to a voice recognition device that has a processor and a memory. The voice recognition device further has: a speaker sample initial value calculation part which extracts an initial speaker sample as an initial value of a speaker sample from voice data having voices of a plurality of speakers mixed on receiving the voice data; a target speaker voice recognition part which performs target speaker recognition based upon the initial speaker sample that the speaker sample initial value calculation part extracts; and a speaker sample update part which extracts a speaker sample again based upon an execution result of the target speaker voice recognition, and updates the initial speaker sample. The target speaker voice recognition part performs the target speaker voice recognition again based upon the speaker sample having been updated by the speaker sample update part.SELECTED DRAWING: Figure 2 【課題】事前に話者サンプルを必要とせずに、複数の話者の音声から音声認識と話者同定を、精度よく実施する。【解決手段】プロセッサとメモリを有する音声認識装置であって、複数の話者の音声が混合された音声データを受け付けて、前記音声データから話者サンプルの初期値として初期話者サンプルを抽出する話者サンプル初期値算出部と、前記話者サンプル初期値算出部が抽出した初期話者サンプルに基づいて目標話者音声認識を実施する目標話者音声認識部と、前記目標話者音声認識の実行結果に基づいて再度話者サンプルを抽出して前記初期話者サンプルを更新する話者サンプル更新部と、を有し、前記目標話者音声認識部は、前記話者サンプル更新部で更新された前記話者サンプルに基づいて前記目標話者音声認識を再度実施する。【選択図】図２