likelihood of the labels in the utts file??

[Pilar O]

Hi Simon, reading “Multisyn: Open-domain unit selection for the Festival speech synthesis system”, I got to this interesting paragraph where it says:

“Other information is associated with individual phones in the linguistic structure to enable the target cost to incorporate a component which deals with suspected bad labelling. This information includes a normalised version of the HMM log likelihood score for each segment and a flag which marks segments which are too short to have meaningful pitch-marking.” (p. 327)

Does this mean that the utts files have for each segment a value of how confident we are that this is the correct label?? Is this used as pruning info?? Is this actually done in the Multisyn version we are using??

Thanks!

[Simon]

Although, in principle, a the likelihoods of the segments could be used as a simple form of confidence measure, this is not actually done in Multisyn.

[Author]

Posts