A native output module based on piper TTS. #996
Conversation
| # single speaker or multi-speaker. Single speaker models produce a | ||
| # single speech dispatcher voice. The speech dispatcher voice name | ||
| # can be listed with 'spd-say -o cxxpiper -L', but it is | ||
| not needed as the voice will be the default, and only, voice available. |
There was a problem hiding this comment.
This line is missing a #
|
|
||
| # ModelPath and ConfigPath are required. There should be exactly one of each of them. | ||
| ModelPath "/home/ddavies/src/piper-models/clean100.onnx" | ||
| ConfigPath "/home/ddavies/src/piper-models/clean100.onnx.json" |
There was a problem hiding this comment.
We cannot hardcode such a path. Is there not standard path defined by piper? Something like /srv/something or /opt/something?
| # file for a multi-speaker model contains a "speaker_id_map" object | ||
| # that lists an integer speaker id and string mneumonic for each | ||
| # speaker supported by the model. Since speech dispatcher has no | ||
| # notion of speaker id, speaker selection detqails are hidden from the |
| # NB: Unsure if onnx models may allow different " languages within | ||
| # the same multi-speaker model. REgardless, if there's sufficient memory: it might be possible to load multiple | ||
| # models and have cxxpiper select between them, while presenting the union of | ||
| # the speakers and languages of each model. |
There was a problem hiding this comment.
Yes, it will be nice long-term to be able to load several models. Even better, list all of them but load them only on-demand, so people don't have to modify any file but just install a model package and see the language pop-up in orca
| # english-us+male7 en-US male7 | ||
| # english-us+whisper en-US whisper | ||
| # english-us+female1 en-US female1 | ||
| ## english-us+croak en-US croak |
| else { | ||
| piperConfig.useESpeak = false; | ||
| } | ||
| cxxpiper::initialize(piperConfig); |
| } | ||
|
|
||
| SPDVoice **module_list_voices(void) | ||
| { |
There was a problem hiding this comment.
There's odd indentation, it is not consistent
| cxxpiper_handle_sound_icon(data); | ||
| break; | ||
| } | ||
| } |
There was a problem hiding this comment.
This double } is odd, better either indent the whole content of the switch, or the content of each case.
| DBG("Sending begin event"); | ||
| module_report_event_begin(); | ||
| (void)cxxpiper::textToAudio(piperConfig, voice, cmdInp, | ||
| audioBuffer, result, audioCallback); |
There was a problem hiding this comment.
Does piper support pipelining? To start sending audio to the server before the whole audio is produced. That would be really important to get good reactivity (but it's fine not to have it for first inclusion)
| switch (msgtype) { | ||
| case SPD_MSGTYPE_CHAR: | ||
| case SPD_MSGTYPE_KEY: | ||
| case SPD_MSGTYPE_SPELL: |
There was a problem hiding this comment.
We shouldn't ignore them, that'd break some screen reading support.
I have added to CI a rule to install piper in |
|
Grateful for your fixes to CI and Makefile.am -- thank you. |
'Native' means it's written in C++, so it does not incur the overhead of the generic output modulebased on the piper command line utility.