摘要
隐马尔可夫模型（ＨＭＭ）技术是语音识别中应用较为成功的算法，但它的缺点影响了其精度、速度、硬件实现和推广应用，神经网络（ＮＮ）具有并行性、强的分类能力和易于硬件实现等优点。将ＮＮ与ＨＭＭ相结合构成混合网络，能克服ＨＭＭ与ＮＮ的缺点，保留双方的优点，本文详细评述了目前在语音识别中应用的由ＨＭＭ和ＮＮ构成的四种混合网络。通过对其结构、识别性能和特点的分析，可以看出ＨＭＭ和ＮＮ构成的混合网的性能明显优于纯ＨＭＭ和ＮＮ，是更适于语音识别的网络。
Hidden Markov Model(HMM)is a successfully used algorithm in speech recognition, yet its disadvantages limit its performance, speed, hardware implementation and application. Neural network (NN)has many advantages such as parallel processing ability, powerful discriminating ability, ease of hardware implementation etc.The hybrid systems based on the combination of HMM and NN can overcome their disadvantages while maintain their advantages.In this paper, four kinds of HMM and NN hybrid systems which have been widely used in speech recognition were described in detail. From the discussion of their structure, recognition performance and characteristics, it can be seen that the gybrid systems are superior to HMM and NN thus are more suitable for speech recognition.
出处
《电子学报》
EI
CAS
CSCD
北大核心
1994年第10期73-80,共8页
Acta Electronica Sinica
关键词
神经网络
隐马尔可夫模型
混合网络
语音识别
Neural networks, Hidden Markov model, Hybrid networks, Speech recognition
