Autoregressive HMM for HTS

This is an implementation of the autoregressive HMM built on top of HTS. It provides the ability to do embedded re-estimation and decision tree clustering for linear-Gaussian autoregressive output distributions.

An example of how to use this implementation is available (autoregressive HMM version of the HTS demo).

Download

This extension is released as a patch file for HTS, which is in turn released as a patch file for HTK. The modifications represented in the patch file below are released under a BSD-style license. However please note that once you apply this patch file to HTS, you must obey the license of HTS and license of HTK.

Prerequisites

• HTK 3.4, available from the HTK software archive (registration required)
• HTS 2.1 for HTK 3.4 patch file, available from the HTS release archive

Current version

• ar-for-hts 0.8 for HTS 2.1 patch file

Old versions are available in the archive.

Relevant publications

• M. Shannon and W. Byrne (2010)
  Autoregressive clustering for HMM speech synthesis
  Proc. Interspeech 2010
  
• M. Shannon and W. Byrne (2009)
  Autoregressive HMMs for speech synthesis
  Proc. Interspeech 2009