Method of speech recognition

A method of speech recognition, comprising the steps of:

• analyzing an input signal every frame and deriving feature parameters from the input signal, the input signal lasting for a period having a part containing input speech;
• generating an input vector from the feature parameters of a plurality of frames;
• forming partial standard patterns to contain a correlation between frames of speech data by using speech data of a plurality of frames;
• calculating partial distances between the input vector and partial standard patterns while shifting the frame one by one, wherein standard patterns correspond to recognition-object words respectively, and each of the standard patterns is composed of the partial standard patterns which represent parts of the corresponding recognition-object word respectively;
• accumulating the partial distances into distances between the input speech and the standard patterns, the distances corresponding to the recognition-object words respectively;
• comparing the distances with each other every frame, and selecting a minimum distance of the distances and one of the recognition-object words which corresponds to the minimum distance every frame;
• comparing the minimum distance in frames preceding the current frame and the minimum distance in the current frame, and selecting a smaller distance of the minimum distance in frames preceding the current frame and the minimum distance in the current frame as a new minimum distance to update and memorize the minimum distance and the selected recognition-object word;
• and
• deciding the memorized recognition-object word which is present at an end of the input signal to be a recognition result.

A method of speech recognition, comprising the steps of:

• analyzing an input signal every frame and deriving feature parameters from the input signal, the input signal lasting for a period having a part containing input speech;
• generating an input vector from the feature parameters of a plurality of frames;
• forming partial standard patterns to contain a correlation between frames of speech data by using speech data of a plurality of frames;
• calculating partial distances between the input vector and partial standard patterns while shifting the frame one by one, wherein standard patterns correspond to recognition-object words respectively, and each of the standard patterns is composed of the partial standard patterns which represent parts of the corresponding recognition-object word respectively;
• accumulating the partial distances into distances between the input speech and the standard patterns, the distances corresponding to the recognition-object words respectively;
• comparing the distances with each other every frame, and selecting a minimum distance of the distances and one of the recognition-object words which corresponds to the minimum distance every frame;
• comparing the minimum distance in frames preceding the current frame and the minimum distance in the current frame, and selecting a smaller distance of the minimum distance in frames preceding the current frame and the minimum distance in the current frame as a new minimum distance to update and memorize the minimum distance and the selected recognition-object word;
• and
• deciding the memorized recognition-object word which is present at an end of the input signal to be a recognition result,
• wherein the partial-distance calculating step comprises using a statistical distance measure based on a posteriori probability in measuring the partial distances.