Re: Raspberry Pi /w Callie
![Post Post](./styles/prosilver/imageset/icon_post_target.gif)
Posted:
Sun Jul 21, 2019 12:26 pm
by xanatos
I currently have all my voices running with minimal delay. The system still has to write to a wav then aplay the wav, but it's running on first phrase after boot about 12 seconds, then subsequent utterances have about a 1 to 1.5 second delay, more obvously for longer sentences.
Note also that a large text dump takes a long time to record to wav, so if you dump sentences at a time vs. paragraphs, it's more responsive.
I'll see if I can dig up my install & setup notes. FYI, all my stuff will be for 3B+ running Stretch.
These are great voices for the Pi if you, like me, wish to run everything offline and not require a network connection as a precondition of operation.
Dave
Re: Raspberry Pi /w Callie
![Post Post](./styles/prosilver/imageset/icon_post_target.gif)
Posted:
Sat May 30, 2020 4:18 am
by orangepeel
BUMP!
Any fixes to this issue that appears to be still outstanding?
oss_audio failed to open
Thanks!
Re: Raspberry Pi /w Callie
![Post Post](./styles/prosilver/imageset/icon_post_target.gif)
Posted:
Mon Jun 01, 2020 7:52 am
by xanatos
The delays have become a minimal issue over time. Running "Belle" on my Raspberry Pi 4 (4Gig) with Buster, the delays are trivial at this point. The sentence "Knowledge is knowing a tomato is a fruit, wisdom is knowing to not put it in a fruit salad" takes less than a second from being passed to the speech engine to the time speech begins.
The main issue that remains for me is the fact that the voices need to write the wav file to the disk, then it must be played back. Because the system uses an SD card, this creates many writes to the card over time. I would like to find an app that could take the wav that would normally be written and instead buffer it to ram and speak it without creating a physical wav file on the card. I would guess this to be a relatively simple task by redirecting the output in they way R Pi users can direct output to stdout, or /dev/null, etc., but I haven't researched this particular function in any detail yet, concentrating instead on cognitive architecture and language processing for now... but it is an issue I will need to revisit in the not-too-distant future.
Glad to see people are still using this thread and these voices. These voices still stand out as among the best available anywhere for embedded, offline, standalone applications. I'm hoping that the developers do continue to add functionality over time, especially some vastly more robust ability to script voice tone, tempo, emphasis, etc. I use the in-line markups but they are only minimally helpful.
Dave