marcelloceschia found the lines in the MaryTTSAudioStream.java, where the sampling rate is hard-coded…
Thanks marcelloceschia for looking into it!
He also proposed a quite easy solution to it, but was unable to build the bundle himself:
Since I’m not able too, I’d like to ask for somebody more experienced in Java programming than me to step in and build some jar according his proposal …
Thank you for testing. I hope it will be tested by some other users to and we can open a pull request du fix this issue in the main branch.
Please also try to switch voices and check the result for all voices, especially the en_US voice with 48kHz sampling.
@curlyel maryTTS has a lot more voices with “other quallity”, but this can result in > 500MB library code
I’ve tested all three voices supplied with maryTTS bundle. The two German voices at 16kHz are rendered correctly as well as the English one at 48kHz. So from my point of view all fine
Regarding the “other quality voices”:
Since nowadays the bundle is just installed on user request, an increased size of the bundle would only affect those users, who decided actively to get maryTTS installed for having TTS local. Those would accept higher disc space requirements - probably
I personally would be more than happy if could have a better TTS quality for the cost of some disc space and compute power. If you’re preparing a PR, it would be nice getting some more voices especially in higher quality in.
@marcelloceschia [quote=“curlyel, post:91, topic:15576”]
I’ve tested all three voices supplied with maryTTS bundle. The two German voices at 16kHz are rendered correctly as well as the English one at 48kHz. So from my point of view all fine
[/quote]
I tested it with (Build #920) version 2.1.0.201705151816 | MaryTTS Voice Service, but both german voices are too fast, and english voice is like a bit on Valium (a bit too slow). I’ve tested it directly on OH2-Server (System Speaker with mp3 support), it’s ok (all voices)!!! But when I select my “Allplay speaker” ->‘Default Sink’ in audio configuration, speed didn’t fit… Is it an Allplay problem or a MaryTTS problem?
@Celaeno1 I can confirm the issue (same problem here), however I cannot see how the could be caused by the AllPlay binding. The binding is just forwarding the generated voice URL to the AllPlay speaker.
It would make sense to analyze the generated file which is sent to the speaker, however it seems like this only exists temporarily, I have no clue how to access it.
@marcelloceschia I don’t know. I used it 1st time after sonos patch was published. Would it be easy for you to provide a JAR file with changed to 16 kHz? My “Ecplise IDE” is too old. I have to update (or reinstall) to new version first! Thanks a lot!
@Celaeno1 you do not need the eclipse ide, yust use mvn to build the package. I can build an package with fix 16kHz sampling rate if you like to test it.
@marcelloceschia Yes, a testing with 16 kHz would be great. I downloaded your JAR from your personal message. And now both German voices are running! English is in slowmotion, but I don’t need it.
@Celaeno1 can we do some tests, to check why the automaticaly desiction does not work? I’d like to add a debug log to see if this.audioFormat is != null