VOICE RECOGNITION METHOD, DEVICE, APPARATUS, AND COMPUTER READABLE STORAGE MEDIUM
To provide a voice recognition method based on double decoding in which acoustic variety of one acoustic model covers a disadvantage that there are few acoustic paths of other acoustic models, two decoding paths are independent from each other, a decoding space is expanded, and accordingly the accur...
Gespeichert in:
Hauptverfasser: | , , |
---|---|
Format: | Patent |
Sprache: | eng ; jpn |
Schlagworte: | |
Online-Zugang: | Volltext bestellen |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | To provide a voice recognition method based on double decoding in which acoustic variety of one acoustic model covers a disadvantage that there are few acoustic paths of other acoustic models, two decoding paths are independent from each other, a decoding space is expanded, and accordingly the accuracy of voice recognition can be improved.SOLUTION: The voice recognition method includes: for an inputted voice signal, obtaining first acoustic decoded information by a first acoustic model, and obtaining second acoustic decoded information by a second acoustic model generated by joint modeling of sound and language; and determining a first set of candidate recognition results based on the first acoustic decoded information, determining a second set of candidate recognition results based on the second acoustic decoded information, and determining a final recognition result for the voice signal based on these two sets of candidate recognition results.SELECTED DRAWING: Figure 2
【課題】1つの音響モデルの音響多様性により他の音響モデルの音響経路が少ないという欠点を補い、2つの復号化経路が互いに独立しており、復号化空間が拡張され、それにより音声認識の正確率が向上されることができる二重復号化に基づく音声認識方法を提供すう。【解決手段】音声認識方法は、入力された音声信号に対して、第1の音響モデルにより第1の音響復号化情報を取得し、音響と言語の組合せモデリングにより生成された第2の音響モデルにより第2の音響復号化情報を取得するステップと、第1の音響復号化情報に基づいて第1組の候補認識結果を確定し、第2の音響復号化情報に基づいて第2組の候補認識結果を確定し、これら2組の候補認識結果に基づいて音声信号に対する最終的な認識結果を確定するステップを含む。【選択図】図2 |
---|