Speech recognition using neural networks - Chapter 8

Comparisons Trong chương này chúng tôi so sánh hiệu suất tốt nhất của chúng tôi NN-HMM lai so với các hệ thống khác nhau, cả hai cơ sở dữ liệu Hội nghị đăng ký và quản lý tài nguyên cơ sở dữ liệu. Những so sánh này cho thấy sự suy yếu tương đối của các mạng lưới dự báo, sức mạnh tương đối của các mạng lưới phân loại, và tầm quan trọng của tối ưu hóa cẩn thận trong bất kỳ phương pháp tiếp cận nhất định | 8. Comparisons In this chapter we compare the performance of our best NN-HMM hybrids against that of various other systems on both the Conference Registration database and the Resource Management database. These comparisons reveal the relative weakness of predictive networks the relative strength of classification networks and the importance of careful optimization in any given approach. . Conference Registration Database Table shows a comparison between several systems all developed by our research group on the Conference Registration database. All of these systems used 40 phoneme models with between 1 and 5 states per phoneme. The systems are as follows HMM- Continuous density Hidden Markov Model with 1 5 or 10 mixture densities per state as described in Section . LPNN Linked Predictive Neural Network Section . HCNN Hidden Control Neural Network Section augmented with context dependent inputs and function word models. LVQ Learned Vector Quantization Section which trains a codebook of quantized vectors for a tied-mixture HMM. TDNN Time Delay Neural Network Section but without temporal integration in the output layer. This may also be called an MLP Section with hierarchical delays. MS-TDNN Multi-State TDNN used for word classification Section . In each experiment we trained on 204 recorded sentences from one speaker mjmt and tested word accuracy on another set or subset of 204 sentences by the same speaker. Perplexity 7 used a word pair grammar derived from and applied to all 204 sentences perplexity 111 used no grammar but limited the vocabulary to the words found in the first three conversations 41 sentences which were used for testing perplexity 402 a used no grammar with the full vocabulary and again tested only the first three conversations 41 sentences perplexity 402 b used no grammar and tested all 204 sentences. The final column gives the word accuracy on the training set for comparison. 147 148 8. Comparisons .