 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Sampling Cantonese software
Recently, commercial Cantonese voice recognition software packages have been marketed in Hong Kong with claims of reducing the effort of inputting Chinese text into computers by providing voice-to-text conversion. But how do they square up?
An Information Technology Branch study team was formed to trial the software and explore the possibility of applying voice-recognition technology to enhance Force efficiency, particularly in the report room setting.
The team found input time was reduced in preparation work for lengthy Chinese documents, including report room statement taking, translation of papers and witness statements and reports.
Complaints Against Police Office staff, Police Translators and Chinese Language Officers, who deal with a lot of Chinese documents, were invited to sample the software in their daily work. Executive Officers, Management Services Officers and Personal Secretary IIs of ITB were also included in the trial to see if the technology could help casual users in preparing Chinese documents.
Currently there are two products for voice recognition of Cantonese dialogue on the Hong Kong market. One is the "Voice Writer" from Pen Power and the other is Speech*Pen*Keyboard Chinese Input Suite from Lernout & Hauspie Asia. The core of the "Voice Writer" package, however, was built on L & H technology.
Copies of the Voice Writer and SPK Chinese Input Suite were passed to staff for a two-month trial. Sample text was used to compare the input rate between keyboard input and voice recognition software. The result showed the average recognition rate for the latter, including error correction, was 0.25 characters per second. The average keyboard input performance for a trained user is around 0.5 - 0.8 characters per second, which is 2 - 3 times faster than the packages' performance.
However, when compared to casual users, the measured performance of keyboard input and speech recognition were roughly the same. At the end of the trial users were interviewed. While the software packages were easy to use, users found two major constraints in using it in report rooms. Firstly, the accuracy of the software in recognising Cantonese speech was not satisfactory (ranging from 50 to 80 per cent).
Secondly, many colloquial phrases, swearing and special characters which are frequently used in court documents were not provided in the software. Considerable effort was required for users to build up their own vocabulary to cater for their operational needs.
However, users agreed that the software would be suitable for use in a normal office setting for production of non-urgent lengthy Chinese documents.
ITB is monitoring the development of this technology and exploring its potential in all possible future applications within the Force.