[mimictts] Add LRU cache

[dalgwen]

Hello,

This pull request add the common TTS cache capability to the mimic TTS Service. (MimicTTSService.java)

And also : [mimictts] Fix playing with audio servlet
It also fixes an issue with a sound not playing with servlet based sink. (AutoDeleteFileAudioStream.java)
(The system responsible for deleting a file that is not used anymore was too greedy. The sound file was deleted, detecting a closing stream, before even being read for "real")

[lolodomo]

Maybe your class AutoDeleteFileAudioStream could be moved into core framework to be used by different add-ons?

[dalgwen]

Hum yes, but is it really a good implementation for core ?
I designed it as a workaround (temporary), and it is not very bug proof. If the stream is not read by the sink, for whatever reason, then it won't even be deleted.

The more I think about it, the more I think it should be handled in the AudioHTTPServer implementation.
Instead of :
String serve(FixedLengthAudioStream stream, int seconds);
The method could be :
String serve(AudioStream stream, int seconds);

And then, in the AudioServlet implementation, the serve (AudioStream stream, int second) method implementation could

• if stream is FixedAudioStream, then as proceed usual (putting the stream in a multi time enabled map for further reference)
• if not, then using a temporary cache file, a wrapper, or whatever magic, and still putting it in the multi time enabled map.

Then we won't have to do workarounds in all TTS services for the audio sinks that only support this multi time serve (FixedLengthAudioStream stream, int second) method.

WDYT ?

[lolodomo]

Looks like a great idea.

[lolodomo]

@dalgwen : do you propose to implement your idea (AudioHTTPServer) ?

[lolodomo]

@dalgwen : I have another idea. synthesizeForCache is called by the cache handler in two cases:

1. When the cache is enabled and a new TTS has to be added in the cache
2. When the cache is disabled

Our problem is only with the second case. Why not distinguishing the two cases and add a new method synthesizeForDisabledCache in AbstractCachedTTSService. For this method, the TTS service would know that it has to create a file and return a FileAudioStream.
mimic and voiceRRS would then build a temporary file when this method is called.

WDYT ?

Edit: In fact, we just need to know in synthesizeForCache if the cache is enabled or disabled. I have to check if this information is accessible. No need for a new method.
Edit2: it is not available. We could add another argument for method synthesizeForCache to requestr a FileAudioStream.

[dalgwen]

do you propose to implement your idea (AudioHTTPServer) ?

I made a try :
dalgwen/openhab-core@9b95051

With this, the servlet and the audiosinks using it do not need to restrain to FixedLengthAudioStream.
The servlet take a copy in memory if it is small (< 1MB, to avoid disk write operation), or in a temporary file (should be very rare because 1MB is, I think, larger than the majority of use cases). It then serves the copy. And the copy can be cloned and used several times.

Of course, if it not needed (if the TTS service or the cache already made a file), there is no modification.

I still have one weird behaviour : the code to delete the file when expired is called, but the file remains on disk.

[lolodomo]

Looks like a good solution? I hope you can fix your weird behaviour with file not deleted.

[dalgwen]

I did a PR for what we discuss here.

In the meantime, I put this PR on "draft" mode

[lolodomo]

@dalgwen : is it ready for a review ?

[dalgwen]

@dalgwen : is it ready for a review ?

I just force pushed after a rebase.

As this PR removes the workaround for the HTTP audio servlet, it will break compatibility with audio sink using audio servlet and still testing against FixedLengthAudioStream.
Apart of this, it is ready.

[lolodomo]

As this PR removes the workaround for the HTTP audio servlet, it will break compatibility with audio sink using audio servlet and still testing against FixedLengthAudioStream.
Apart of this, it is ready.

Please unmark the PR as draft in this case.
I will merge it as soon as most of audio sinks are changed.

[lolodomo]

LGTM, thank you

[lolodomo]

I will wait a little (maybe end of weekend), I hope someone will merge my changes in audio sinks.
But this PR will be included in Milestone 5.