MLLRMaximum-Likelihood Linear Regression
MLLRMaster of Labour Law and Relations (Australia)
Copyright 1988-2018, All rights reserved.
References in periodicals archive ?
When adaptation was performed using the MLLR or MAP technique, the error rate was considerably reduced.
Variance compensation within the MLLR framework for robust speech recognition and speaker adaptation.
(70) AO MLLR, File 7-14-0-90, box 3, Ontario Legislative Assembly, 14 February 1951, 3.
(73) AO MLLR, File 7-14-0-93, box 3, Regulations Under The Hours of Work and Vacations With Pay Act, 1944, 6 July 1944.
AO MLLR, File 7-14-0-130, box 3, The Minimum Wage Act, Order No.
(88) AO MLLR, File 7-14-0-90, box 3, Globe and Mail, "Labour Delegates Clash on Official Coddling," 13 January 1951.
We tried to determine the appropriate number of transformation matrices, or clusters, for MLLR. MLLR with seven clusters (silence, consonants, and five Japanese vowels) achieved the best performance in the experiment.
Maximum likelihood linear regression (MLLR) [30, 31] was the adaptation technique used for the native ASR-baseline system in order to make it usable for non-native speakers.
MLLR is based on the assumption that a set of linear transformations can be used to reduce the mismatch between an initial acoustic model set and the adaptation data.
When the user writes his/her name in the field "Escribe Nombre de Usuario" the interface updates the list of the pop-up menu "Selecciona Usuario" and creates the directories to store the MLLR transformations.
After all sentences are recorded the user just needs to press "Realiza Adaptacion" (Perform Adaptation) to perform MLLR adaptation of the HMMs of the baseline ASR.
When doing this, the MLLR transformations and directories for that user are loaded.