TQE: Transcription Quality Evaluation

1
TQE Transcription Quality Evaluation

• A CLARIN-NL project

Radboud University Nijmegen Institute for Dutch
Lexicology Max Planck Institute for
Psycholinguistics
2
TQE practical information

• Duration 01/04/2010 01/07/2011
• Type Demonstrator Project
• Project team
• CLST Centre for Language and Speech Technology
• Helmer Strik (coord.), Joost van Doremalen,
  Eric Sanders,
• Catia Cucchiarini, Robin Oostrum, Ferdy Hubers
• INL Instituut voor Nederlandse Lexicologie
• Remco van Veenendaal, Laura van Eerten
• MPI Max Planck Institute for Psycholinguistics
• Daan Broeder, Tobias van Valkenhoef, Peter
  Withers
• CLARIN centre
• MPI Max Planck Institute for Psycholinguistics
• Daan Broeder

3
(No Transcript)
4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
Automatic TranscriptionQuality Evaluation

• Input
• Audio signals
• Phone(tic) transcriptions
• Output
• For each phone TQE measure
• How
• Audio and phonetic transcriptions are aligned
• Phone boundaries are derived
• For each phone a TQE measure is determined,
• a confidence measure, e.g. ranging from 0-100
• indicating how well phone segment fit
  together,
• i.e. what the quality of the transcription is

8
MPI version
9
CLST development version
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
Survey 2a) De bestandsformaten
Antwoord Telling
WAV 30 34,88
OGG 6 6,98
AIFF 13 15,12
MP3 16 18,60
MP4 5 5,81
FLAC 5 5,81
ALAW 4 4,65
ULAW 3 3,49
anders 4 4,65
17
Survey 2c) De opnameprecisie
Antwoord Telling
8 bit 3 7,89
12 bit 3 7,89
16 bit 24 63,16
24 bit 8 21,05
18
Survey 3) De formaten en standaarden voor
fonetische transcripties
Antwoord Telling
SAMPA 23 28,40
X-SAMPA 6 7,41
IPA 25 30,86
CGN-set 9 11,11
YAPA 3 3,70
Celex 7 8,64
LH 3 3,70
anders 5 6,17
19
Survey 4) De software
Antwoord Telling
Praat 32 53,33
Audacity 10 16,67
CoolEdit 7 11,67
Audition 5 8,33
anders 6 10,00
20
Survey 8) Interesse in opname CLARIN-infrastructu
ur
Antwoord Telling
Ja 22 64,71
Nee 11 32,35
Weet niet 1 2,94
21
Survey 9) Bereid tot meeleveren metadata
Antwoord Telling
Ja 27 90
Nee 3 10
22
Survey 10) Huidig gebruik van metadataforma(a)t(e
n)
Antwoord Telling
OLAC 1 2,38
IMDI 4 9,52
CMDI 5 11,9
Dublin Core 4 9,52
TEI 4 9,52
Geen 21 50
Anders 3 7,14
23
Epilogue

• More information
• http//lands.let.ru.nl/strik/research/TQE/
• Questions ?

THE END
24
(No Transcript)