University of Minnesota DuluthSearch | People | Departments 

Utilizing Computer Dictation for Language Sample Transcripts

Mark Mizuko & Rudolph Chmelik

University of Minnesota Duluth

WebNet 2001, Association for the Advancement of Computing in Education

Session 383

Introduction 

Language samples are one of the best methods speech-language pathologists have to measure a client's language proficiency. Language samples are gathered from the client, recorded and then transcribed into a visual format. When compiling language samples, recordings of the client's speech must be made, which typically consist of 50 to 100 communication units (utterances). The first step in attaining a language sample is to record a client's speech. The next step lies in the transcription and analysis of the recorded language sample. The compilation of the language sample is an incredibly time consuming process; listening to a recording of the client's speech sample and manually logging it onto hard copy, either handwritten or word processed. With normal speaking rates at about 170 words per minute, the process of typing this sample out involves a great deal of listening, pausing, and typing. As time consuming as this transcription process is, a method to expedite the transfer of data into a format more easily dealt with would be a welcome thing.

Purpose

The purpose of this study was to study the effectiveness of using a computer dictation program, Dragon Naturally Speaking (DNS), to expedite the language sample analysis procedure. DNS is a voice recognition program that transcribes verbal output into written text. The feasibility of using DNS in the clinical environment depends on two factors; 1) the ability of the program to accurately transcribe a client's speech, and 2) to do so with a minimal amount of individual configuration.

Method

A speech sample was recorded and manually transcribed (see attached, Form IA). The researcher then began the initialization sequence required of DNS. After installation and the first training session, the written sample was then verbally input to the computer using DNS, utilizing the researchers voice. This sample was read twenty five trials. The number of errors was tallied for each trial, and percentage of accuracy were calculated. This procedure was then repeated four more twenty-four trials. A note on DNS: DNS is a voice transcription program which is designed to learn the speech patterns of a specific individual and accurately record them. Initially, a user must complete a brief training session with DNS. A training session involves the user to read a written text (chosen by DNS) out loud to the computer. The principle DNS revolves around is at the more training sessions a user completes, the more accurate DNS will be at transcribing their speech . The version used for this research was Dragon Naturally Speaking Preferred USB, Version 5.0.A note on the computer: DNS was operated on a PC under Windows 98. The relevant technical specifications for the PC are as follows: Pentium II 350Mhz processor with 384 MB of PC lOO - z SDRAM.

Results

Table 1 

Sample

T1
T2
T3
T4
T5
Mean Error Rate
Accuracy Rate
Training Time
Cumulative Time
1
30
32
32
36
38
33.6
93.55%
5.08 min
5.08 min
2
34
30
29
32
32
31.4
93.97%
10.21min
15.29 min
3
30
30
31
28
29
29.6
94.32%
4.01min
19.3 min
4
25
26
25
25
26
25.4
95.12%
23.25 min
42.55 min
5
25
23
25
22
23
23.5
95.49%
24.28 min
67.23 min

 

Discussion

It took 23 minutes and 6 seconds to manually transcribe this language sample with 100% accuracy. It took 5 minutes and 1 seconds (5 minutes and 8 seconds training time not included in this measurement) to read sample 1 to the computer to achieve 93.55% accuracy. Over the course of the five training sessions with Dragon Naturally Speaking, it took 67 minutes and 23 seconds to read the sample to the computer to achieve 95.59% accuracy. The difference and gain between the first and twenty-fifth trial was 62 minutes, 15 seconds with a gain of 1.94% in accuracy.

The potential for this method to save time is difficult to dismiss. The margin of error exists, but only to a small degree. If this procedure were to be used in a clinical setting, the benefits of it's use would easily outweigh any difficulties associated with it (i.e. learning to use the program, training with DNS, and the margin of error). Overall, it seems that this technique is a beneficial time saving method which busy speech pathologists would find useful.


Form IA - Manual Transcription

This story is called october sky

but I don't know why

there's a kid like say

he was like 16 years old and

he played for football and he real horrible at it but then

he got home everybody's on the yard cause

there's a satellite up in outer space going past his house

when he saw it then the next day made him interested to make rockets

little ones.

so he'd launch them off

his first one he was right in the back yard front his back yard

it took him a while to build it and

when he light it he had some friends over then

when they're backing up when he turns it

you think this is ok to get really close from it until

the rocket exploded part of their gate broke until

their mom said do that somewhere else but not where in your neighborhood so but then then

his dad work in the mine he's like the

he's the owner of it

he's the boss and so

they were working behind it behind the behind his mine talk about

behind his office couple feet away from it well

there's a metal thing that can go behind it but

when they light it they went behind the metal behind a wall it was all metal until

the rocket then it went up in the air just like

its going like straight but

it Uke a jet fast cut guys apart half his finger off

the the guy just stand there like until the rocket went in the garbage can

his dad comes out and tells his son to go somewhere else far far away from town and from his mine and

he found the rocket then he threw it back in the garbage again throw away until then then

they going far & away like seven mile away

he just goes by himself until his friend drop over

they'll help carry all the stuff for him until

they riding on a truck and back of their pickup someone else's

when they went to the mile they found a place quiet place to go until

at his dads workers one of the workers was like one of his friends sons

was not son but was not like a kid but was like 36 years old and like

they hang around each other sometimes and

he guys a mechanic and he can do

he's like a blacksmith he told

he asked him to help make a like a little fort thing like there's

a little small one they can go in and and when they launch rocket they can protect them

at nighttime they're stealing like logs at a log place they were stealing some and

after they did that then they build like a real small cabin and

that's the fort for them and so

their first rocket they launch over there there's there's

a brown guy brown brown people and his son went and

there was a dad

when they launch launch it

their dad the kids dad was right by the drivers door so

 

Time to transcribe: 23:06 # of Utterances: SO # of Words: 521

 

Go Back to Previous Page

 

[an error occurred while processing this directive]

© 2014 University of Minnesota Duluth

The University of Minnesota is an equal opportunity educator and employer.

Contact Us | Privacy

Last modified on 04/09/04 01:53 PM

Page Coordinator: Mark Mizuko, mmizuko@d.umn.edu

CSD Home

 

Times since December 7, 2001 this page has been accessed: 131962