Fourteen of the acoustic models were trained using the TC-STAR data described before, and they consist of gender-independent and gender-dependent acoustic models obtained from different partitions of the data.
The behavior of combining characters from different partitions was evaluated by examining the relative contribution, or utility, of data partitions to resolving relationships of the combined data set.
The behavior of combining characters from different partitions was evaluated by examining the relative contribution, or utility, of data partitions to resolving relationships within the simultaneous analysis tree by calculating the Partitioned Bremer support (PBS) using TreeRot3 [ 24].
Normalizing data from different ways of partitioning the same population will provide incompatible conclusions about the associations that hold in the total population.
Classical partitioning techniques such as hash, range partitioning are simple to use, but they result in distributed transactions when accessing data from different servers.
There is probably much information in common between resources generated from the different partitions.
We stress that relative rate tests, likelihood ratio tests, taxon deletion experiments and comparisons of trees generated from different data partitions and using different methods cannot prove LBA, but they are suggestive of this explanation (or, at very least, consistent with it).
Different partitions from this data set were used for calculating divergence times: ABV (1+2) = concatenation of first and second codon positions of the three genes (3421 sites); A2BV2 = concatenation of first and second codon positions of ADRA2B and VWF plus all codon positions of BRCA1 (4350 sites); and ABV (1+2+3) = concatenation of all codon positions of the three genes (5130 sites).
Estimates of natural levels of rate variation ranged from 0.05-3.38 for different partitions.
15 different partitions into data sample and 'oracle' data set were generated to average random fluctuations.
