Description: 基于GMM的概率神经网络PNN具有良好的泛化能力，快速的学习能力，易于在线更新,并具有统计学的贝叶斯估计理论基础，已成为一种解决像说话人识别、文字识别、医疗图像识别、卫星云图识别等许多实际困难分类问题的很有效的工具。而且PNN不但具有GMM的大部分优点，还具有许多GMM没有的优点，如强鲁棒性，需要更少的训练语料，可以和其他网络其他理论无缝整合等。-GMM based probabilistic neural network PNN good generalization ability, the ability to learn fast, easy online updates, and with the Bayesian statistical theory based on estimates, and has become a solution as speaker recognition, text recognition, medical image recognition, satellite images and other real recognition when difficulties classification of very effective tool. But GMM PNN is not only the most advantages, but also has many advantages GMM not as strong robustness, require less training corpus, and other networks to other theories, such as seamless integration. Platform: |
Size: 7158 |
Author:姜正茂 |
Hits:
Description: 基于GMM的概率神经网络PNN具有良好的泛化能力，快速的学习能力，易于在线更新,并具有统计学的贝叶斯估计理论基础，已成为一种解决像说话人识别、文字识别、医疗图像识别、卫星云图识别等许多实际困难分类问题的很有效的工具。而且PNN不但具有GMM的大部分优点，还具有许多GMM没有的优点，如强鲁棒性，需要更少的训练语料，可以和其他网络其他理论无缝整合等。-GMM based probabilistic neural network PNN good generalization ability, the ability to learn fast, easy online updates, and with the Bayesian statistical theory based on estimates, and has become a solution as speaker recognition, text recognition, medical image recognition, satellite images and other real recognition when difficulties classification of very effective tool. But GMM PNN is not only the most advantages, but also has many advantages GMM not as strong robustness, require less training corpus, and other networks to other theories, such as seamless integration. Platform: |
Size: 7168 |
Author:姜正茂 |
Hits:
Description: 常用的说话人识别方法有模板匹配法、统计建模法、联接主义法(即人工神经网络实现)。考虑到数据量、实时性以及识别率的问题，采用基于矢量量化和隐马尔可夫模型(HMM)相结合的方法。
说话人识别的系统主要由语音特征矢量提取单元(前端处理)、训练单元、识别单元和后处理单元组成，
-Commonly used methods of speaker recognition template matching method, statistical modeling method, and connection method (ie, artificial neural networks). Taking into account the amount of data, real-time as well as the recognition rate of the problem, based on vector quantization and Hidden Markov Model (HMM) method of combining. Speaker recognition system mainly by the voice feature vector extraction unit (front-end treatment), training modules, identification and post-treatment unit modules, Platform: |
Size: 64512 |
Author:孙丽 |
Hits:
Search - neural networks speaker recognition