查詢結果分析
來源資料
頁籤選單縮合
| 題 名 | 智慧型手機平臺下語音辨認之強健式信賴度分析與模型補償技術探討=An Investigation on Robust Confidence Measure and Model Compensations for Smartphone-based Speech Recognition |
|---|---|
| 作 者 | 洪維廷; | 書刊名 | 資訊科技國際期刊 |
| 卷 期 | 6:1 2012.06[民101.06] |
| 頁 次 | 頁32-38 |
| 分類號 | 448.84 |
| 關鍵詞 | 強健式語音辨認; 信賴度分析; 模型調適; Robust speech recognition; Confidence measure; Model compensation; |
| 語 文 | 中文(Chinese) |
| 中文摘要 | 本文針對PDA/智慧型手機運算資源受限的平台,提出以HMM為基礎之語音辨認核心的強健式技術。我們主要探討的問題如下: (1) 如何在資源受限的移動式裝置與雜訊環境下進行信賴度分析 (2)如何在資源受限的移動式裝置下進行模型補償,進而達到效能提升的目的。整個設計流程經歷過演算法修改、定點數分析、移植到PDA/智慧型手機驗證等程序,最終將相關程式模組移植到Dopod 818Pro 進行實機測試。由於本文提出的設計能不需耗用過多的資源下達到一定程度的抗雜訊能力,咸信非常適合作為運算資源受限的平台之語音辨認核心。 |
| 英文摘要 | This paper presents a structural design for robust HMM-based speech recognition system under PDA/smart-phone platforms. Two main issues were investigated: (1) how to implement the CM (confidence measure) under noisy environment with resource-constrained platform? (2) how to perform model compensation for improving recognition rate under noisy environment with resource-constrained platform? We have gone through algorithm modification, fixed-point analysis and porting procedures to PDA/smart-phone. The final testing is performed with the Dopod 818pro machine. As the proposed design of robust modules need not consume many computing resources, it is conceivably suitable for speech recognition of resource-constrained platforms. |
本系統中英文摘要資訊取自各篇刊載內容。