移动平台上的DSR：Java ME与Symbian的比较
需积分: 0 176 浏览量
更新于2024-11-03
收藏 1.68MB PDF 举报
"Java与Symbian：手机上软件实现的分布式语音识别对比"
这篇论文主要探讨了在当今具有强大处理能力的移动电话上，如何合理利用分布式语音识别（DSR）技术。作者Dmitry Zaykovskiy和Alexander Schmitt来自德国乌尔姆大学的信息技术研究所，他们对比了两种主流的移动操作系统——Java ME和Symbian上DSR前端标准的软件实现。
首先，文章介绍了ETSI（欧洲电信标准化协会）的DSR前端标准，并在Java ME和Symbian平台上进行了软件实现的性能比较。这一部分可能涉及了各种语音处理算法、数据结构以及系统优化策略，以确保在不同平台上的高效运行。
接着，论文讨论了执行时间这一关键指标，这是衡量DSR在实际设备上是否能实时运行的重要因素。通过对当前移动设备的测试，作者证明了在这些设备上进行实时特征提取是可行的。特征提取是DSR过程中的重要步骤，它涉及到对语音信号的预处理，如滤波、分帧、加窗、傅立叶变换等，以提取出有助于识别的关键信息。
然而，论文也揭示了在DSR开发和部署过程中可能遇到的挑战和难题。这些可能包括但不限于内存限制、计算资源的优化、平台兼容性问题、语音数据库的管理和更新，以及由于手机硬件限制导致的识别率下降。特别是在词汇量增加时，由于内置的低复杂度模型和有限的处理能力，识别率可能会显著降低。
此外，关键词还提到了“特征提取”和“分布式语音识别”，这表明论文深入研究了这两个领域。特征提取不仅关乎算法选择，还可能涉及到特征维度的压缩、特征选择以及特征学习等技术，以减少计算负担并提高识别准确率。而分布式语音识别则关注如何在多个设备或服务器之间分布计算任务，以实现大规模、高效的语音处理。
这篇论文对于理解Java ME和Symbian平台在DSR技术上的差异和实际应用提供了深入的见解，对于移动设备上的语音识别开发者和研究者具有很高的参考价值。它提醒我们在设计和实施DSR解决方案时，必须充分考虑设备的硬件限制和实时性需求，同时解决开发和部署过程中可能出现的各种问题。
Java vs. Symbian: A Comparison
of
Software-based DSR
Implementations on Mobile Phones
Dmitry Zaykovskiy and Alexander Schmitt
Institute
of
Information Technology, University
ofUlm,
Ulm, Germany
{dmitry.zaykovskiy,
alexander.schmitt}@uni-ulm.de
Abstract
With increasing processing power
of
today's mobile
phones, a reasonable employment
of
Distributed Speech
Recognition (DSR) Technology comes into reach. This pa-
per presents the ETSI DSR front-end standards as software-
based implementations on the two most popular mobile
phone platforms: Java ME and Symbian.
We
present per-
formance comparisons between the different front-end stan-
dards as well as the execution times on current mobile
phones. After showing that real-time feature-extraction on
current devices is possible, we furthermore point out hin-
drances and pitfalls during development and deployment.
Keywords: Java ME, Symbian, mobile phones, feature
extraction, distributed speech recognition (DSR).
also restricted to a very limited quantity
of
words. With a
growing number
of
entries in the phone book, the recog-
nition rate suffers severely due to built-in low-cost proces-
sors, limited storage and RAM. With this embedded strat-
egy applications such as SMS dictation or the use
of
nat-
ural language are far out
of
reach for the near future. The
most vividly discussed proposal to overcome this challenge
is the principle
of
Distributed Speech Recognition. In this
approach, the speech recognition process is separated into
two parts: a
front-end on the client-side and a back-end on
the server-side.
The front-end extracts characteristic features out
of
the
speech signal, whereas the back-end, making use
of
the lan-
guage and acoustic models performs the computationally
costly recognition.
1 Introduction
Figure
1:
Architecture
of
a client-server DSR system.
2 Related Work
Figure 1 shows a system architecture for DSR. The client
captures the speech signal using a microphone and extracts
features out
of
the signal. The features are compressed in
order to obtain low data rates and transmitted to the server.
At the server back-end, the features are decompressed and
subjected to the actual recognition process.
Bit-Stream Decoding
and
Error Mitigation
Bit-stream
Server
Client
Recognit~on
Result:
While there have been a number
of
studies on the the-
ory
of
DSR, little research has been done on deploying
of
this technology to real mobile devices. The authors
of
[1] proposed a modified version
of
the widespread ETSI
FE-standard for DSR, implemented as a hardware-based
front-end solution for a Motorola Digital Signal Processor
(DSP). A software-based implementation
of
the ETSI AFE-
standard for PDAs using Sphinx-4 [10] as speech recog-
• the number
of
recognizable words is very limited,
• usually the words have to be recorded by the user be-
forehand,
• the recognition system
is
speaker dependent.
Some recent devices spare the user the necessity to pre-
record the commands and feature speaker-independent, em-
bedded speech recognition. This functionality, however, is
The days are numbered where we used our mobile phones
exclusively for telephone conversation. Today we have ac-
cess to thousands
of
different applications and services for
our mobile companions and their number is rapidly grow-
ing. Although the devices have stopped getting smaller and
smaller, we see ourselves confronted with a limited user
interface consisting
of
tiny keys and a miniature display,
which suffices for making phone calls, but which is un-
suited to control applications. The most promising solution
to this challenge
is
the use
of
speech recognition.
If
we take a glance at modem mobile phones, we indeed
discover basic speech recognition functionality: most mo-
bile devices on the market support voice control features
such as voice dialling or hands-free commands. Instead
of
searching names in the telephone book, the user can dictate
the name
of
a person he wants
to
call or, on some devices,
use a voice command to launch a particularphone function.
Although this technology points out new ways to improve
user interfaces on mobile phones, it still has severallimita-
tions:
Authorized licensed use limited to: IEEE Xplore. Downloaded on March 17, 2009 at 15:53 from IEEE Xplore. Restrictions apply.
下载后可阅读完整内容，剩余5页未读，立即下载
点击了解资源详情
点击了解资源详情
点击了解资源详情
2023-06-02 上传
2021-06-29 上传
2021-06-29 上传
2023-06-04 上传
2023-12-21 上传
2023-12-21 上传
chen715660403
- 粉丝: 0
- 资源: 4
我的内容管理
展开
最新资源
- Raspberry Pi OpenCL驱动程序安装与QEMU仿真指南
- Apache RocketMQ Go客户端：全面支持与消息处理功能
- WStage平台：无线传感器网络阶段数据交互技术
- 基于Java SpringBoot和微信小程序的ssm智能仓储系统开发
- CorrectMe项目：自动更正与建议API的开发与应用
- IdeaBiz请求处理程序JAVA：自动化API调用与令牌管理
- 墨西哥面包店研讨会：介绍关键业绩指标(KPI)与评估标准
- 2014年Android音乐播放器源码学习分享
- CleverRecyclerView扩展库：滑动效果与特性增强
- 利用Python和SURF特征识别斑点猫图像
- Wurpr开源PHP MySQL包装器：安全易用且高效
- Scratch少儿编程：Kanon妹系闹钟音效素材包
- 食品分享社交应用的开发教程与功能介绍
- Cookies by lfj.io: 浏览数据智能管理与同步工具
- 掌握SSH框架与SpringMVC Hibernate集成教程
- C语言实现FFT算法及互相关性能优化指南
