Mbrola tool

[marcteyssier]

Hi,

First of all, thank you for your work. I was using SNIP, but since I learned that the service was going to close, I installed Rhasspy and it was very easy to use.

I see a tools named Mbrola, link to Espeaker to add intonation on the robotic voice and the result is very nice. https://github.com/numediart/MBROLA

Is it possible to integrate it in Rhasspy ? Or to use it ?

Regards Marc

[djusHa]

Hi,

thanks for the tip, it sounds really good, but what we need is the -s speed option for espeak ;-)

Good news, it can be done. Mbrola implementation in rhasspy is very easy. Bad news, you must build your own docker image.

Maybe i can make a pull request @synesthesiam ?

[synesthesiam]

This looks like a great tool. Is this in any way related to SSML?

To get started, one easy way to integrate this into Rhasspy would be through the Command TTS system. That simply calls an external program with the text string on standard in and expects WAV data on standard out.

but what we need is the -s speed option for espeak ;-)

I've added an "arguments" option to the espeak profile config in the (upcoming) Rhasspy 2.4.18 :)

[djusHa]

If mbrola binary is already exist in /usr/bin and mbrola voice file is in /usr/share/mbrola/de5 for Example, it can be used like so (i've added the speed option myself):

In Docker file i'm using this to compile mbrola.

# compile mbrola
ARG mbrolaURI=https://github.com/numediart/MBROLA.git
WORKDIR /tmp
RUN mkdir mbrola && \
    mkdir -p /usr/share/mbrola && \
    cd mbrola && \
    git clone $mbrolaURI . && \
    make && \
    cp Bin/mbrola /usr/bin && \
    cd .. && \
    rm -rf mbrola

[synesthesiam]

Work on Rhasspy's text to speech support has moved to https://github.com/rhasspy/rhasspy-tts-cli-hermes