Finalize the Adapted Acoustic Model

After IDOL Speech Server processes the adaptation data set, you have one or more accumulator files, depending on whether you split the adaptation data set into smaller chunks during the training processes. The final stage in adapting an acoustic model is to apply all the accumulator files to the original model.

To finalize the adapted acoustic model

  1. Create a list that contains the file names (excluding file extensions and paths) of all accumulator files (see Data Naming Scheme). Create a list even if you have only one accumulator file.

    For more information about IDOL Speech Server's list manager, see Create and Manage Lists.

  2. Send an AddTask action to IDOL Speech Server, and set the following parameters:

    Type The task name. Set to AmTrainFinal.
    InAm The acoustic model file to adapt.
    AccList

    The list that specifies the accumulator files.

    AccPath The path to the directory that contains the accumulator files.
    OutAm The name of the adapted acoustic model to create.

For example:

http://localhost:13000/action=AmTrainFinal&InAm=C:\LP\ENUK\ver-ENUK-5.0-16k.am&AccList=ListManager/accList&AccPath=C:\AMTRAIN\acc&OutAm=myModel.am

This action uses port 13000 to instruct IDOL Speech Server, which is located on the local machine, to apply the accumulators stored in C:\AMTRAIN\acc to the ver-ENUK-5.0-16k acoustic model to produce the adapted acoustic model myModel.am.

This action returns a token. You can use the token to:

You can set the following parameters to further configure the adaptation operation.

Relevance

How much relevance to give to the original model parameters when updating the means and variances of the models. This effectively assigns a weight in frames to the original values, as if this number of frames had been seen during adaptation. The higher this number, the more examples are needed in the adaptation data for the model to change significantly.

If you are running the AmTrain task in rapid adaptation mode, HPE recommends that you set the Relevance parameter to 0 (zero). Any other value limits the effectiveness of the adaptation process.

MinEgs The minimum number of examples (frames) that have to have been seen for a specific model parameter to be updated.

For more information about the [amadaptend] module parameters, see the IDOL Speech Server Reference.


_HP_HTML5_bannerTitle.htm