查詢結果分析
來源資料
相關文獻
- Speech Recognition Based on Multi-Processor System
- An Overview of RNN-Based Mandarin Speech Recognition Approaches
- Combining Two-dimension Cepstrum and Extended Back-propagation Algorithm to Speech Recognition
- 利用類神經網路來處理與語者無關之中文辨識
- GSM語音編碼演算法之改良
- 類神經網路中文語音聽寫系統
- 快速學習類神經網路手寫英數與中文字辨認系統
- A Large-Vocabulary Speech Recognition System for Taiwanese (Min-Nan)
- 應用類神經網路之語音切割
- 文字轉語音系統之一般性架構
頁籤選單縮合
題名 | Speech Recognition Based on Multi-Processor System=多處理機在類神經網路語音辨認系統上之應用 |
---|---|
作者 | 蔡永昌; 蔡玉娟; 劉榮宜; 游世賢; 陳玲芬; |
期刊 | 電信研究 |
出版日期 | 19930800 |
卷期 | 23:4 1993.08[民82.08] |
頁次 | 頁527-536 |
分類號 | 448.71 |
語文 | eng |
關鍵詞 | 多處理機; 語音; 辨認系統; 類神經網路; |
中文摘要 | 本篇文章主要在探討如何利用多重處理機以縮短在類神經網路學習時間,並以學習式向量量化(LVQ)為骨幹並結合傳統之K-Means演算法,以求得類神經網路之初始值,而非以亂數當成每個神經單元之初始值。在學習過程中隨著時間之變化並作加溫後冷却(annealing)方法,使得類神經單元上之值能趨近最佳化。 本系統硬體架構主機為個人電腦(486-PC),及傳算器(包含十顆三十二位元中央處理單元,及20M RAM, I/O速率為10Mbit/Sec),整個合成為一環狀架構,而由PC作整個系統之控制。 |
英文摘要 | In this paper, we present an new, efficient and structured clustering algorithm for pattern classification and also implement this algorithm on a multi-processor system. This algorithm is modified by merging Lloyd’s algorithm (K-Means) [1], Learning Vector Quantization (LVQ) [2, 3, 4] algorithm and annealing technique. In a pattern classification system, LVQ algorithm is used as main algorithm for vector quantization, where K-Means algorithm is used in in-class training as a preprocessor, and annealing technique is applied in training phase of this system. From our experimental results, they show that of LVQ algorithm or LVQ algorithm with annealing technique or LVQ algorithm combined with K-means algorithm without using annealing technique. The success of our algorithm has been demonstrated via simulation on 1140 isolated digit Mandarin Mono-Syllables data. The data is collected from 76 peoples. We have achieved satisfactory recognition rate 99.73% which is better than that of TDNN 97.8, and HMM 97.1 (These two algorithms are also running on our multi-processor system). |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。