The latest version of Dragon NaturallySpeaking promised increased accuracy and enhancements such as support for e-mail and Internet browsing. The Preferred Edition has a massive word vocabulary of about 230,000 words with about 30,000 active words. It had the ability to create your own voice macros and playback your dictation in various synthesised voice tones.
System Requirements
We recommend that a system with Pentium III 500 MHz or higher should be used, otherwise there is a lot of noticeable lag and the system suffers long bouts of intense processing. Using the software along with Microsoft Word made the system very sluggish and you require more processing and RAM power if you plan to run e-mail and Internet applications in conjunction with the software.
Installation
Installation was pretty basic and it also went through the usual validation of the microphone volume and silence detection routines. The microphone volume settings were done automatically and background noise cancellation too was taken care of.
Integration
Most importantly, Dragon NaturallySpeaking is highly compatible with any word-based application, making it very useful for people who have various free word-based applications. It is also compatible with a wide variety of e-mail clients such as Microsoft Outlook, Lotus Notes, Eudora, etc.
Training
The initial training period lasted for about 5 to 7 minutes. It typically leads you through an entire essay that you have to read out. The software then builds a voice file according to the profile and then builds on the profile as you use the software more often.
Accuracy
Dragon NaturallySpeaking scored very well with consistent scores of 90 per cent to 95 per cent accuracy all through the dictation sequence.
+ Fast training procedure
+ Very accurate
+ Indianised to a certain extent
- Expensive
Specifications:
Minimum System requirements: Windows 98/Me/2000 or Windows NT 4.0 (with SP-6 or greater), 266 MHz Intel Pentium processor with MMX or equivalent, 128 MB RAM (higher recommended for greater accuracy), 150 MB free disk space, soundcard (SB16 compatible),
speakers.
Features: Noise cancelling headset and high-quality microphone were included along with the package.
Benchmarks: 90 per cent to 95 per cent accuracy, 60 per cent to 70 per cent CPU utilisation during dictation (pretty high).
Contact: HCL Infosystems
Phone: 011-4526518
E-mail: 011-4526579
Price: Rs 17,995
Rating: A+
==============================================
Philips Free Speech 2000
Get a mike along
The Philips Free Speech 2000 is built around IBM Via Voice speech recognition engine. It allows dictation in 12 other languages and also a voice command interface. FreeSpeech 2000 uses a specified 290,000-word background database, which was developed with Oxford University Press.
FreeSpeech 2000 came along with its very own SpeechMike. The SpeechMike unit can be used much like a regular TV remote or a microphone.
System Requirements
The software was definitely low on system resources and there was little lag while speaking into the microphone at normal talking pace. While working with Microsoft Word, the system did become a little sluggish, but it was within tolerable limits.
Installation
The installation process was simple, and the installation of the speech engine was fast too. The onboard microphone was rather disappointing as it took more than four tries to get the acceptable quality level. We also tested the software without the SpeechMike, using a generic headset. In this case it yielded slightly better results because of the constant distance of 1-inch that the microphone stays from the mouth while using the headset.
Integration
FreeSpeech 2000 too has been designed to work with a wide variety of applications such as Microsoft Office, Corel suites and other e-mail clients.
Training
The initial training period lasted for about 7 minutes but the interface was pretty clumsy.
Accuracy
The accuracy was around 70 per cent to 80 per cent in the first round, though the second time around the accuracy jumped up to over 85 per cent. Third time around the accuracy was up to an acceptable level of 92 per cent.
+ Comes with an integrated trackball with the SpeechMike
+ Low on system resources
- Expensive
Specifications:
Minimum system requirements: Intel Pentium 166 MHz with MMX (Pentium II recommended)
Windows 9x/NT 4.0, 48 MB RAM (64 MB when dictating in Microsoft Word), Sound Blaster compatible soundcard, speakers
Features: Includes the very well designed SpeechMike with integrated track ball and left and right click buttons, microphone and a speaker.
Benchmarks: 70 per cent to 92 per cent accuracy, 50 per cent CPU utilisation during dictation (quite low).
Contact: Philips India
Phone: 022-6912351
Fax: 022-6912499
E-mail: sameer.dania@philips.com
Price: Rs 17,995, Rs 15,995 (without the SpeechMike but with a headset)
Rating: A
===============================================
VoiceDirect Continuous Gold
Bronze in performance
IMSI's VoiceDirect Continuous Gold also uses IBM's Via Voice Technology. Using a base vocabulary of 22,000 words which is expandable and customisable to 64,000 words, it also has a 260,000 backup word dictionary. This software too features the voice command feature but along with the IMSI text to speech module, which makes the computer do the reading from the text. Another great feature of this software is the Dictation Macro Editor, which is very simple to use and creates a macro that carries out a series of actions.
System Requirements
The software was really light on the system and used very little resources. Though the minimum software requirements are pretty low, it is highly recommended to use at least 64 MB RAM while using Microsoft Word in conjunction with this software.
Installation
The installation was simple, but had a tiresome registration menu. However, this can prove useful for software updates. This was the only software package which detected the soundcard used. Moreover, the audio quality and volume level setup were excellent.
Integration
Includes support for the usual word-based applications, Microsoft applications along with several organizers including Schedule+.
Training
The program initiates the User Wizard to collect information about the user, starts the microphone setup program and initialises a typical sentence reading program to improve the initial speech-recognition accuracy and performance. More importantly, the continuous feedback given by the software with regard to pausing between words was intuitive.
Accuracy
Accuracy with the initial 5 minute training was not enough and only after repeated sessions did the software reach acceptable levels of accuracy. By the third round, the accuracy just came up to around 85 per cent, which is just about acceptable. The training too was pretty frustrating since certain parts had to be gone over and over again.
+ Relatively cheaper
+ Includes an excellent noise cancelling microphone
Specifications:
Minimum system requirements: Intel Pentium 166 MHz with MMX (Pentium II recommended)
Windows 9x/NT 4.0, 32 MB RAM for Windows 9x, 48 MB for Win NT, 125 MB hard disk space. Sound Blaster compatible soundcard or Mwave audio system.
Features: Includes a ready reckoner for all the punctuation and formatting and general commands too. It also comes with a headset with noise cancelling features. You also get a six-month free subscription to updates via the Internet.
Benchmarks: Around 85 per cent accuracy after repeated training, maximum 60 per cent CPU utilisation during dictation (quite low).
Using the Microsoft Speech recognition engine and incorporating a decent 60,000-word vocabulary, it proved to be quite a decent product. It was very simple and basic in the design of its interface, but the training program was definitely one of the easier ones that we came across. The software comes along with the We Tech Dictation Pad which basically is your regular dictation pad. The software is definitely a great bargain for its price. The Indianisation of the software is evident from the fact that the user can now set his own language to translate into English. This is very useful for users having little knowledge of English-they can set standard sentences in their own language and get it typed in English by not actually speaking English to the software. The software also has an automatic voice response system, which allows you to generate a specific Voice response to the Inquiry made by telephone calls to your place.
System Requirements
The software used very little CPU and system resources. It was also observed that the application was very soundcard dependant and a lower specification soundcard would cause the software to lose some of its accuracy.
Installation
The hardware setup was very user friendly. But the initial setup was very different, since all three programs, We Tech text reader, We Tech Dictation pad and the We Tech Voice Command speech codes had to be installed separately using the environment buttons provided on the setup screen.
Integration
You have to use the We Tech dictation pad to dictate; it lacks support for speaking directly into applications.
Training
The initial training period is definitely not enough for starting off with continuous speech recognition. With at least three to four training sessions you can reach some acceptable accuracy levels.
Accuracy
Just with the initial training the accuracy was way below the acceptable limit. Slight variations in voice would cause the software to detect the wrong word. Accuracy initially ranged in the low 50 per cent bracket. However, with more training it did pick up to acceptable levels. But technical jargon was simply too much for it at the first attempt.
+ Very cheap
+ Indianisation to a great extent
- Not very accurate, requires extensive training
Specifications:
Minimum system requirements: Pentium 250 or faster, 64 MB RAM, soundcard or in-built sound hardware, close-talk or handset microphone, Windows 9x/NT
Features: The voice command module was definitely the better part of the package. Very much recommend for people with disabilities.
Benchmarks: 50 per cent accuracy after first training, improved to 75 per cent after repeated training, below 50 per cent CPU utilisation during dictation (quite low).
