# 英文報告 Application ## Voice recognition - Feature extraction - It is the first step.The action create informative and clear data.The data is called feature.Feature is used to training the acoustic model. - Acoustic model - It is the second step. It is the kernal of voice recognition. It use the feature to training the model that can recognize the voice. The main acoustic model is Hidden Markov model currently. - Decode - It is the third step. And it has two ways to decode. - Speech model - If a sentence has similar pronunciation. but the meaning of sentence are far away. For example "We live in a different area" and "We live in a different era".If it is hard to determine the meaning by the feature. Speech model will use context to do it with calculating the probability. - Dictionary - It's function is to map the pronunciation and word. If it hard to map. It will use speech model to help it. - Recognition - It is final step. It is the output for the recognition. - Difficulties - Noise prevention - It is necessary to prevent the noise. Because noise will interfere the Feature extraction. If feature is fill with noise. The result's recognition is wrong. - Recognition of Speech model - It is the hard to do but it is necessary. Speech model use the meaning of context to determine the meaning of sentence. The field include linguistics, psychology and physiology. ## Development of image recognition - Text recognition - OCR is a application of Text recognition. One of example is License plate recognition. You can find it in parking lot. - Digital Image processing & recognition - Digital image use pixel to store the data. Using pixel is easy to compress the data and avoid distortion when sending the data. - Object recognition - image classification - The function is that determing whether the object is in the image. - object localization & object detection - Marking the object's place is the same of two ways. The diffence between object localization and object detection is that object localization is focus on a object, and object detection can detect more than one object.