Arch Linux User Repository

One can subvert the long pauses by feeding piper-tts all the data at once, instead of chunked, as speech-dispatcher by default does for workload reasons.

Add these two lines to your piper.conf

# as adviced by https://htmlpreview.github.io/?https://github.com/brailcom/speechd/blob/master/doc/speech-dispatcher.html#Configuration-files-of-output-modules
GenericDelimiters "|"  # "|" never occurs
GenericMaxChunkLength 99999

I don't think the reasons for having chunking applies to piper-tts, so perhaps this should just be set as the default that this package ships with.

For me this .conf works better - https://github.com/ken107/read-aloud/issues/375#issuecomment-1937517761

Anyone having issue of 'Long pause between Sentences'? - https://github.com/Elleo/pied/issues/9, any way to resolve this?

No longer works

Some problem with onnxrutime.

echo 'Welcome to the world of speech synthesis!' | piper-tts --model /usr/share/piper-voices/en/en_GB/alan/low/en_GB-alan-low.onnx --output_file welcome.wav
terminate called after throwing an instance of 'Ort::Exception'
  what():  Load model from /usr/share/piper-voices/en/en_GB/alan/low/en_GB-alan-low.onnx failed:Protobuf parsing failed.

@jouven I've added these files to piper-voices-common and its "install" file

Best tts for real time text, if hardware allows it, then again it's the fastest one with high quality audio.
My setup:
For ~/.config/speech-dispatcher/speechd.conf add and leave only this AddModule line uncommented:

AddModule "piper-tts-generic"   "sd_generic"    "piper-tts-generic.conf"

Create ~/.config/speech-dispatcher/modules/piper-tts-generic.conf with:

GenericExecuteSynth "echo \'$DATA\' | piper-tts -q -m \"/somemodeldirectory/en_US-libritts-high.onnx\" -s 21 -f - | mpv --volume=80 --no-terminal --keep-open=no -"
#GenericExecuteSynth "echo \'$DATA\' | piper-tts -q -m \"/somemodeldirectory/en_US-libritts-high.onnx\" -s 21 --output-raw | aplay -r 22050 -f S16_LE -t raw -"
AddVoice "en" "male1"   "kal16"

However how to prevent the variation, without making it sound robotic? I'm talking --noise_scale, by default the value of this setting, If I spam the same text, there is some randomness about how it sounds each time but sometimes the "same" delivery sounds noticeably wrong, but if use --noise_scale 0 it always sounds robotic.
@Edit, this last issue has to do with the model, other models don't have this issue.

Thank you for adding a link to the armv7h binary.

I've created a new AUR packages with piper voices:

https://aur.archlinux.org/packages/piper-voices-minimal (1 model, ~120MB)

https://aur.archlinux.org/pkgbase/piper-voices (one split package per locale, ~6GB for all locales)

Can you add something similar to the PKGBUILD?

optdepends=("piper-voices: voices for all languages")

Hi, can you also add support for armv7h?

Thank you for the package, I really love piper. I am using it regularly now having upgraded to version 2023.9.27-1. Firefox is reading German and English news aloud using piper-tts via speech dispatcher. The setup was a bit difficult. I made three files: ~/.config/speech-dispatcher/speechd.conf

AddModule "piper-generic" "sd_generic" "piper-generic.conf"
DefaultModule piper-generic
DefaultVoiceType  "MALE1"
DefaultLanguage "en-US"

~/.config/speech-dispatcher/modules/piper-generic.conf

Debug “1”

GenericExecuteSynth \
"env DATA="\"$DATA\"" VOICE="\"$VOICE\" " RATE="\"$RATE\"" \
 /opt/piper-tts/piper-pipe"

GenericCmdDependency "piper-tts"

AddVoice "de"    "MALE1"    "de_DE-thorsten-medium"
AddVoice "de"    "FEMALE1"  "de_DE-kerstin-low"
AddVoice "en-us" "MALE1"    "en_US-lessac-medium"
AddVoice "en-gb" "FEMALE1"  "en_GB-jenny_dioco-medium"
AddVoice "en-us" "FEMALE2"  "en_US-amy-medium"
AddVoice "en-us" "FEMALE3"  "ljspeech-high"

DefaultVoice “de_DE-thorsten-medium”

and executable /opt/piper-tts/piper-pipe

#!/bin/bash

VOICE_PATH="/opt/piper-tts/voices"

if [[ ${VOICE: -3} = low ]]; then
  ADJ_RATE=16000
else
  ADJ_RATE=22050
fi
ADJ_RATE=$((${RATE::-3} * 30 + $ADJ_RATE))
echo "$DATA" | piper-tts --model $VOICE_PATH/$VOICE.onnx  --output-raw | \
aplay -r $ADJ_RATE -f S16_LE -t raw -

wait

My setup can accommodate the different qualities of piper’s speech models. The high ones cause a bit of latency on my 6 year old core-i3 notebook. Additionally I can turn down speed for foreign languages. Fortunately aplay used pulseaudio as default device. I hope that this helps others to get started.