Nippon Telegraph and Telephone Corporation

Ntt Docomo, Inc.

Sadaoki Furui

Sadaoki Furui and Ntt Docomo, Inc.

Toshiaki Sugimura

Tomoko Matsui

Zhipeng Zhang

Tsutomu Horikoshi

Kiyotaka Otsuji

Yasuhiro Minami

An object of the present invention is to enable optimal clustering for many types of noise data and to improve the accuracy of estimation of a speech model sequence of input speech. Noise is added to speech in accordance with noise-to-signal ratio conditions to generate noise-added speech (step S), the mean value of speech cepstral is subtracted from the generated, noise-added speech (step), a Gaussian distribution model of each piece of noise-added speech is created (step S), the likelihoods of the pieces of noise-added speech are calculated to generate a likelihood matrix (step S) to obtain a clustering result. An optimum model is selected (step S) and linear transformation is performed to provide a maximized likelihood (step S). Because noise-added speech is consistently used both in clustering and model learning, clustering for many types of noise data and an accurate estimation of a speech model sequence can be achieved.

Noise adaptation system of speech model, noise adaptation method, and noise adaptation program for speech recognition

An object of the present invention is to facilitate dealing with noisy speech with varying SNR and save calculation costs by generating a speech model with a single-tree-structure and using the model for speech recognition. Every piece of noise data stored in a noise database is used under every SNR condition to calculate the distance between all noise models with the SNR conditions and the noise-added speech is clustered. Based on the result of the clustering, a single-tree-structure model space into which the noise and SNR are integrated is generated (steps Sto S). At a noise extraction step (step S), inputted noisy speech to be recognized is analyzed to extract a feature parameter string and the likelihoods of HMMs are compared one another to select an optimum model from the tree-structure noisy speech model space (step S). Linear transformation is applied to the selected noisy speech model space so that the likelihood is maximized (step S).

In a speaker adaptation method for speech models, input speech is transformed to a feature parameter sequence like a cepstral sequence, and N model sequences of maximum likelihood for the feature parameter sequence are extracted from speaker-independent speech HMMs by an N-best hypothesis extraction method. The extracted model sequences are each provisionally adapted to maximize its likelihood for the feature parameter sequence of the input speech while changing the HMM parameters of each sequence, and that one of the provisionally adapted model sequences which has the maximum likelihood for the feature parameter sequence of the input speech is selected and speech models of the selected sequence are provided as adapted HMMs of the speaker to be recognized.

Method for speaker adaptation of speech models recognition scheme using

Noise-resistant speech HMMs are composed by: recording noise in the environment of utterance (S.sub.1); preparing HMMs of the noise (S.sub.2); transforming the output probability distribution of each of the noise HMMs and speech HMMs prepared from speech unaffected by noise and multiplicative distortion to a linear spectral domain (S.sub.31); multiplying the speech HMM distribution in the linear spectral domain by a multiplicative distortion W that is an unknown variable (S.sub.321); convoluting the multiplied value and the noise HMM distribution in the linear spectral domain (S.sub.322); inversely transforming the convoluted value to the original domain of the speech HMM (S.sub.33) to compose incomplete noise-resistant speech HMMs each containing multiplicative distortion as an unknown variable (S.sub.3); calculating the likelihoods of the incomplete noise-resistant speech HMMs for input speech and estimating the multiplicative distortion of that one of the incomplete noise-resistant HMMs which has the maximum likelihood (S.sub.4); and substituting the estimated value into the incomplete noise-resistant speech HMMs (S.sub.5).

Growing community of inventors

Tokyo, Japan

Sadaoki Furui