Hello HPR, this is your host Archer72, for another episode of Hacker Public Radio.
My subject today is Piper voice synthesis, continued.

In response to Ken's comment on my show, hpr4172 Re: hpr4072 Piper voice
synthesis, I am responding to his comment with a solution. I'm glad that Ken
commented, because I had put the problem on the back burner and forgot about it.
Both of us had the make command for the Piper github repo fail at 22%.

I ignored it for the time being and compiled Piper for the Raspberry Pi instead.
Now here is the comment that got me to figure out how to get Piper working on my
own Fedora 40 laptop.

Comment #1 posted on 2024-07-24 14:04:27 by Ken Fallon
Fails on Fedora 40

[ 22%] Linking C shared library libespeak-ng.so /usr/bin/ld:
../ucd-tools/libucd.a(case.c.o): relocation R_X86_64_32S against `.rodata' can
not be used when making a shared object; recompile with -fPIC /usr/bin/ld:
failed to set dynamic section sizes: bad value collect2: error: ld returned 1
exit status [snip] make: *** [Makefile:5: all] Error 2

Comment #2 posted on 2024-07-26 09:38:17 by Archer72
Re:Fails on Fedora 40

Hi Ken, I get the same failure, and found that there was a release in 2023

Github Piper repo

I put the downloaded piper directory in /opt , along with the
piper-voices/en directory and have a successful voice output.

uname -r

6.9.5-200.fc40.x86_64

[piper] [info] Loaded voice in 0.33 second(s)
[piper] [info] Initialized piper Output directory: /home/mark/./output.wav

End of comments

In further conversations with Ken, I found that a vital part of Piper voice
synthesis had been forgotten. For example, if you are using Piper to convert a
text file to .wav, the command need to include the following:
input.txt piper executable location, i.e. /opt/piper/piper

--model, and the model location, i.e. /opt/piper-voices/

note, the voice used need to include the voice in .onnx format and also the
voice configuration in .json format

--output_file output.wav

The final script is included in the show notes.

#!/bin/bash
cat "$1" | /opt/piper/piper --model /opt/piper-voices/en_US/kusal/medium/en_US-kusal-medium.onnx --output_file output.wav

That's it! Now if you had downloaded the voices from hpr4172, there should be a
successful voice to text output.

To put the final touches on 'my' voice, which is the en_US-kusal-medium voice,
processed just a bit with the sox program.

## ~/bin/make-my-voice.sh
#!/bin/bash
# Add 2 seconds of silence to the beginning of the file
sox $1 output.wav pad 2

## Reduce clipping
sox output.wav output-mid.wav vol 0.99

# Reduce the tempo by 12%
sox output-mid.wav final_output.wav tempo 0.88

One last thing. It was brought to my attention that the Piper voice, Bryce 'may'
sound like William Shatner in the original Star Trek tv series.

I will put that clip in here and see what the community thinks. Feel free to
leave a comment saying 'yay' or 'nay' on this opinion.

If this tool works for you, feel free to leave comments on this show. Better
yet, record a show of your own. Looking forward to hearing from the
next host, whether it be by text to speech, or a microphone.

Remember to support free software and apps in the F-Droid store if you use Android.

This has been your host Archer72; Bye