Pimatic-TTS (Text to Speech) Released

---

Inspired by @Tim who uses his Sonos speakers for spoken alerts (Really awesome dude ) I have created a new plugin baptised pimatic-tts.

This plugin provides a device and a rule action allowing your Pimatic hardware platform to locally speak out text strings in your own language using the Google TTS API. The device holds parameters such as language, volume, repeats etc. Defining an action in a rule is easy: say "Goodmorning, the coffee is ready" using my-tts-device

Please read the Plugin page before installing as it requires some prerequisites to be satisfied. Also note this is an early beta version, which may have bugs I have not encountered.

Pico2Wave as a local TTS engine has now been integrated as well.

Thoughts and suggestions are welcome, even if it just helps fill a roadmap!

This will be a huge improvement of pimatic and an importand step towards smart home and automation.

If I get you right you want to implement text to speech on the server side? On the server side it is not required to get google involved. There are fairly good packages for offline synthesis.

@mwittig
Yes, the plugin would facilitate TTS on the server side - Although I did read in older threads there is a desire for client side TTS, I am not sufficiently skilled in client side programming unfortunately…

The prototype currently provides an action in the format Say "text with $pimatic-vars" using <language> speed <int> repeat <int> interval <int>

re offline synthesis packages: Good suggestion!
If available packages are good to use I could attempt to implement those as well. In such case I would suggest to opt for an abstraction layer approach where you can configure your offline package and parameters of choice. What is your recommendation?

@dan One thing to look at is “say”. On Raspberry the output is of mediocre quality. Once of the voices sounds a bit light HAL 9000 from the 60s which might be seen as a feature. On other Windows and OSX it is way better as it can utilize the speech synthesis provide with the OS. The current version requires node 6 or needs to be babelized. The last known version to work with node 4 is 0.12

Another thing to look at is pico2wav, which is a speech synthesizer used for Android but it also runs on Raspbian. There are some node.js bindings for that, but I am not sure which one is a good choice.

Please read the plugin page before installing!. Bugs can be logged at the github page: https://github.com/SenTzu01/pimatic-tts/issues

Thanks for this new plugin!
Maybe you could implement a cache of already tts’ified strings in order to lower number of request to Google servers…

@yves911 said in Pimatic-TTS (Text to Speech) Released:

Thanks for this new plugin!
Maybe you could implement a cache of already tts’ified strings in order to lower number of request to Google servers…

Good point (Although I wouldn’t be concerned with the amount of requests sent to the allmighty Google) !
Only messages with variables will need to be freshly converted. Static messages can be cached / saved indefinitely. Having a local repository also provides some resilience in cases where internet connectivity is disrupted.
Did think of this functionality, just wanted to get the initial version out as quickly as possible, and introduce functionality piece-meal, rather than waiting for a month to have a big-bang.

Current priority is to integrate Pico2Wav (offline TTS). This already yields code which can be re-used for the caching mechanism used with online resources
Caching could follow soon after!

If you have some time, I wouldn’t mind if you would log feature requests on Github as well. This helps me track and prioritize!

Thanks I created a short ticket on GitHub for that.

pimatic-tts v0.0.2 is available.
New Feature:

• Support for Pico2Wave TTS added (as per discussion with @mwittig)
  • Speech resources for text strings are cached on the filesystem (/tmp by default).
  • Text strings including Pimatic variables will not be cached due to their volatile nature.

Pico2Wave is an offline TTS package which can be installed with the package manager (aptitude on Debian systems such as raspbian). The number of languages is limited, but does include EN, DE, FR, IT, ES.

Thoughts / Suggestions for feature enhancements are welcome!
Next priority is to implement caching of cloud based Speech resources.

Please discuss features on the plugin page. I appreciate defects being logged (Alongside forum discussion) on Github

Hey Dan,
don’t really now if I would declare it as failure, therefore I write it here. I today installed your plugin. Love it btw…

After restart an error appear [pimatic] error: Could not initialize the plugin "tts": Cannot find module './pico-device-config-schema'

I checked the plugin folder and really the file was missing. After downloading it from GitHub and put it on site everything was fine. No idea if it is only an local problem for me or something is missing in the download package.

@ortin Nope, you found my sloppiness. its missing from package.json! I’ll fix and release asap.
Just uploaded v0.0.3, which should include the file at install

@tim Just had a chat with @leader21 re TTS.
If I understood correctly you are well experienced in the front-end? My skills in this area are insufficient. However there seems to be a desire in the community to bring TTS to the front-end as well. Just wanted to open a dialog to see if you would be interested in taking on some of the effort for the front-end. I am building my knowledge on integrating TTS systems while developing, and you seem to have done some cool stuff in this area in the past. Just PM me if you are open to a chat Rotterdam style

@Dan. Works now thx!

Feature request:
Could you implement that loudness can be give as attribute within a rule like:

when something happens then say "something" using google-tts with vol 30

With this the loudness is variabel for daily or chill times.

One question regarding the speech velocity of goolge tts. Even on 100 the good girl sounds a little dreamy. I think this is a google limitation and can’t be speed up little bit, right?

@ortin Agree to implement. Most of the required logic is still in place. Please log a feature request at Github for this.

Caching Cloud based TTS resources as per @Yves911 is now first on the list. yours is to follow straight after.
Re speech volume: I have fiddled around with google quite a bit, finding speed 40 to be quite allright for the Dutch voice. Not sure how this relates to other languages really. Let me perform some more tests. I have found the coding dependencies between the various audio streams to be quite unforgiving.

Version 0.0.4 available - New Feature: Static TTS resources are now cached on the local filesystem

@yves911 said in Pimatic-TTS (Text to Speech) Released:

Thanks for this new plugin!
Maybe you could implement a cache of already tts’ified strings in order to lower number of request to Google servers…

@ortin said in Pimatic-TTS (Text to Speech) Released:

Feature request:
Could you implement that loudness can be give as attribute within a rule like:

when something happens then say "something" using google-tts with vol 30

@OrTiN and all: Implemented in version 0.0.5 with a slightly different rule syntax. Volume, Repeat, and Interval can now optionally be parameterized per rule. For parameters not provided parameter values from the device config will be used instead.

when something happens then
<[say | speak]> "something really $useful" <using [tts_device_id | tts device name]> [[with volume nn] [repeating n times] [every n seconds]]

@dan said in Pimatic-TTS (Text to Speech) Released:

Version 0.0.4 available - New Feature: Static TTS resources are now cached on the local filesystem

@yves911 said in Pimatic-TTS (Text to Speech) Released:

Thanks for this new plugin!
Maybe you could implement a cache of already tts’ified strings in order to lower number of request to Google servers…

Hi, just tried and it works very nicely I will certainly replace my own implementation of the Google API by yours
Again thanks

@Yves911 Thanks for the feedback, much appreciated!

For the record: Pimatic-tts doesn’t do FS housekeeping tasks (yet). So small chance the cache could fill the partition up eventually if you are already low of free space (think size of an average speech resource is around 50k maybe less).
If I am bored I might implement something, but IMHO the system owner should run a tight ship

@yves911 said in Pimatic-TTS (Text to Speech) Released:

Hi, just tried and it works very nicely I will certainly replace my own implementation of the Google API by yours
Again thanks

Brief headsup: Development on tts is ongoing. Currently working on integration with UPNP media player devices on the network. Obviously this will greatly improve functionality for TTS. Initial tests against my samsung smart TV are succesful, but I have found some vendors do not completely adhere to the DLNA spec, or the NPM modules have not been written correctly. Rightnow considering whether to fix the dependencies, or just write the full stack myself…