Piper as speech-dispatcher plugin

[takov751]

I was able to hack it together, it was rather simple.
Install speech-dispatcher then in
/etc/speech-dispatcher/modules/piper.conf

GenericExecuteSynth "echo \'$DATA\' | piper --model /path/to/voice/${VOICE} --output_raw | pw-play --rate 22050 --channel-map LE - "
AddVoice "en-GB" "male1" "en/en_GB/alan/medium/en_GB-alan-medium.onnx"
AddVoice "en-GB" "female1" "en/en_GB/jenny_dioco/medium/en_GB-jenny_dioco-medium.onnx"
AddVoice "en-US" "male1" "en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx"

/etc/speech-dispatcher/speechd.conf

AddModule "piper" "sd_generic" "piper.conf"
DefaultVoiceType  "male1"
DefaultLanguage   en-GB
DefaultModule   piper

and you have piper as narrator on your desktop. Of course there is lot of room for improvement, but slowly getting there
/path/to/voice/ should be the directory you have downloaded the voices from huggingface

[lamyergeier]

Is it possible to prevent piper from terminating and restarting after every sentence, it takes a lot of time after each sentence?

I saw this in the log of speech-dispatcher.

[FlippedBit01]

Hey @takov751 this works pretty well for being "hacked together". I've just encountered two issues so far. The first is that the audio stops if I write a long sentence using spd-say, so it doesn't finish saying it. And the other is that apps that connect to speech-dispatcher like firefox and foliate give me an error saying they can't use speech-dispatcher.

Does anyone know how these issues can be resolved?

[FlippedBit01]

Nevermind, after a few reboots it's working fine
. _.)

[takov751]

That is interesting, thanks for reporting back. I forgot to add different language models so that it can be addressed by other apps even if it's hardcoded

[omega3]

Could you please give some description to non-technical users like me:

AddVoice "en" "MALE1" 

DefaultVoiceType  "MALE1"

what these values are and how can I replace them to chose different voice? How to find this classification for equivalent of "MALE1", for example
libritts_r medium 8699(1)
or jenny_dioco
or en_GB-northern_english_male-medium.onnx?

I typed piper --help and don't see any --list-voices command. I downloaded them from hugging face and so far applied from command line pointing to onnx file.

For example in Plasma Okular there is an option to change voice (this also sometimes means language). But with proposed configuration I don't know how to make other voices available for speech dispatcher.

[takov751]

You are making a very good point. my example was hard coded and did not calculated this in. I have modified my first post so that it accommodates those specific voice options. this way you are able to add even different languages as well. and you will be able to choose between the voices with just sdp-say -l en-GB -t male1 and okular should be able to work. I was unable to test changes above yet myself.

[Jopp-gh]

I want to use another voice(german) instead the default one(english):
spd-say -o piper -l de-DE "Hallo alle zusammen"

not sure why I still hear the default english voice. My code should work and use the german voice I've installed. If I change the default voice under pied's settings to german, all english articles sound wrong, of course. (OS:Mint).

[takov751]

Could you show us your /etc/speech-dispatcher/modules/piper.conf ?

[Jopp-gh]

Yes, sure!
there's no /etc/speech-dispatcher/modules/piper.conf file under root, only in my user path, under ~/.config/speech-dispatcher/modules/piper.conf.
I added that one as attachment and changed my user name to username. The en voice in use is called "Ryan"

piper.conf_035321.txt

[takov751]

Yes, sure!

there's no /etc/speech-dispatcher/modules/piper.conf file under root, only in my user path, under ~/.config/speech-dispatcher/modules/piper.conf.

I added that one as attachment and changed my user name to username. The en voice in use is called "Ryan"

piper.conf_035321.txt

So i think in your case the issue is that VOICE parameter is hardcoded.
Based on this:

GenericExecuteSynth "echo \'$DATA\' | piper --model /path/to/voice/${VOICE} --output_raw | pw-play --rate 22050 --channel-map LE - "
AddVoice "en-GB" "male1" "en/en_GB/alan/medium/en_GB-alan-medium.onnx"
AddVoice "en-GB" "female1" "en/en_GB/jenny_dioco/medium/en_GB-jenny_dioco-medium.onnx"
AddVoice "en-US" "male1" "en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx"

Your file should look like this

...

 echo \'$DATA\' | /home/username/.var/app/com.mikeasoft.pied/data/pied/piper/piper --model /home/username/.var/app/com.mikeasoft.pied/data/pied/models/${VOICE} --output_raw | $PROCESS | $OUTPUT;"


AddVoice "en-us" "MALE1" "en_US-ryan-high.onnx"
AddVoice "de-DE" "MALE1" "de_DE-karlsson-low.onnx"

This way you will have optional voice model

[omega3]

You only showed piper.conf. Check speechd.conf. In the link I provided above you have examples. My contains:

AddModule "piper" "sd_generic" "piper.conf"
DefaultVoiceType  "FEMALE1"
DefaultLanguage   en
DefaultModule   piper

Timeout 30
LogLevel  2
LogDir  "default"

DefaultVolume 100
DefaultVoiceType "FEMALE1"
DefaultLanguage "en"
DefaultPunctuationMode "some"

SymbolsPreproc "char"
SymbolsPreprocFile "gender-neutral.dic"
SymbolsPreprocFile "font-variants.dic"
SymbolsPreprocFile "symbols.dic"
SymbolsPreprocFile "emojis.dic"
SymbolsPreprocFile "orca.dic"
SymbolsPreprocFile "orca-chars.dic"

DefaultCapLetRecognition  "none"
DefaultSpelling  Off

AudioOutputMethod "pulse"
AudioPulseDevice "default"
AudioPulseMinLength 10

AddModule "piper"                   "sd_generic"   "piper.conf"

DefaultModule "piper"

LanguageDefaultModule "en"  "piper"
LanguageDefaultModule "pl"  "piper"

Include "clients/*.conf"

And the command you provided works for me for both languages. Also in E-book viewer I have all voices available but in Dolphin I can have only English.

But if you use Pied just move all configurations elsewhere and Pied should create both config files. It will recreate them every time you change Pied settings.

[Jopp-gh]

You only showed piper.conf. Check speechd.conf.

I did not include the contents of speechd.conf because I forgot about but checking speechd.conf doesn't change anything, as most settings in there are commented out and there isn't any setting that could change / influence Pied's configuration. Basically, it is a very generic file.

And the command you provided works for me for both languages

I'm glad they worked for you

But if you use Pied just move all configurations elsewhere and Pied should create both config files

yes, I thought the same, best will be to save both configurations somewhere and rewrite the main piper.conf file every time via shell. Pretty hacky solution but doable.

I tried both Takov's and your configurations correcting my user path, but the terminal uses always the same default voice set in Pied.
Meanwhile I also informed the author of Pied, the current implementation is not flexible enough.

[Cameo007]

Hi, I followed this guide (I think) but it doesn't work.
en_GB-alan-medium.onnx & en_GB-alan-medium.onnx.json are located at /rust/hificnnct/tts-models/en/en_GB/alan/medium/.

When I try this command (just for testing), it works.

echo "hello everybody!" | piper --model /rust/hificnnct/tts-models/en/en_GB/alan/medium/en_GB-alan-medium.onnx --output_raw | pw-play --rate 22050 --channel-map LE -

But the integration with speechd doesn't.

Here is my config
~/.config/speech-dispatcher/speechd.conf

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

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

GenericExecuteSynth "echo \'$DATA\' | piper --model /rust/hificnnct/tts-models/${VOICE} --output_raw | pw-play --rate 22050 --channel-map LE - "
AddVoice "en-GB" "MALE1" "en/en_GB/alan/medium/en_GB-alan-medium.onnx"

In the logs I can see

/run/user/1000/speech-dispatcher/log/speech-dispatcher.log

[Mon May 27 16:01:42 2024 : 178445] speechd:    Module set parameters.
[Mon May 27 16:01:42 2024 : 178819] speechd:  Error: Module reported error in request from speechd (code 3xx): 300-Opening sound device failed. Reason: Cannot open plugin server. error: file not found.
300 MODULE ERROR

/run/user/1000/speech-dispatcher/log/piper.log

terminate called after throwing an instance of 'std::runtime_error'
  what():  Model config doesn't exist

It seems that piper cannot find the model's config even though it is at the right place.

[takov751]

Your suggestion doesn't help. And can the files be corrupted when it works when using the exact same command (of course with ${VOICE} replaced) in the terminal?

Interesting. I have no idea at this point. Right now i have very similar setup and mine works when i specify language( for some reason it does not work without it) like spd-say -l en "asd". What distro are you using and does it use pipewire?

[Cameo007]

I'm on NixOS 24.05 with Pipewire enabled. If you are interested in it and you know Nix, my config is public (https://codeberg.org/Cameo007/NixOS-Configs)

[takov751]

I'm on NixOS 24.05 with Pipewire enabled. If you are interested in it and you know Nix, my config is public (https://codeberg.org/Cameo007/NixOS-Configs)

I am familiar with nix and i have a suspicion that it could be a speech-dispatcher packaging issue or at least some sort of access issue.

[Cameo007]

Nice

[vitodb]

@Cameo007 I had the same issue.
I solved it by replacing

--model /path/to/${VOICE}

with

--model /path/to/\'$VOICE\'

[JJJHolscher]

On arch installing

extra/speech-dispatcher # duh
aur/piper-tts
aur/piper-tts-common
# maybe aur/piper-tts-minimal

and rebooting was enough for it to work.

It creates a
/etc/speech-dispatcher/modules/piper-generic.conf
that contains

Debug "1"

GenericPunctNone ""
GenericPunctSome "--punct=\"()[]{};:\""
GenericPunctMost "--punct=\"()[]{};:\""
GenericPunctAll "--punct"

GenericExecuteSynth "env DATA="\"$DATA\"" VOICE="\"$VOICE\"" /usr/bin/piper-dispatcher"

GenericCmdDependency "piper-tts"

AddVoice "en" "MALE1"    "en/en_US/ryan/high/en_US-ryan-high"

DefaultVoice             "en/en_US/ryan/high/en_US-ryan-high"

which for some reason isn't explicitly loaded into the speechd.conf, but is recognized by speech-dispatcher nonetheless.
Do with it what you will.

[krei-se]

i have a blind user running ubuntu (and well actually i cannot install arch there) and wonder what else the package contains. Is there f.e. a place for the models that piper will always find? rn i use a subpath of /etc/speech-dispatcher