htsvoice

htsvoice files are voice models produced and used by the [HTS] [vocal-synthesis] framework. They are generated by training tagged input audio from a speaker/singer and can be used with HTS and tools built on top of it to generate speech or singing waveforms.

Unfortunately, this does not seem to be a widely used or available format. There is one widely available model (hts-voice-nitech-jp-atr503-m001) and a few (literally 2-3) other sites online that I found with others available for free download.

some free htsvoice files I've found

• https://github.com/icn-lab/htsvoice-tohoku-f01
• MMDAgent, a (partially?) open source virtual agent with audio and visual synthesis. Contains the open source "Mei" htsvoice model: https://sourceforge.net/projects/mmdagent/files/MMDAgent/MMDAgent-1.7/MMDAgent-1.7.zip/download

random links and resources

• Preparing Data for Training an HTS Voice

comparison to other voicebank formats

HTS voices are models that are trained, which according to someone on the [utau] forums requires "around 4 hours of english recordings to do a decent HMM voice"