Notices

Hmm, openrepos still distributes 2.1-3, have you published 2.2 somewhere else?

...I always forget to hit "save" after uploading a new version to OpenRepos. Thanks for pointing that out.

On my Jolla, I can't get Saera to work. She keeps on 'Initializing speech recognition...' forever.
Installed it last week for the first time. Installed packages:
portaudio 19_20140130-0
eSpeak for sailfish 1.46.02-10.6.1.jolla
Saera 2.2-0
What can I do to test or to provide more info?

I remember something about empty contacts or/and at least one music track caused that. But it should have been fixed?

Update:

Up til now, I have been doing development solely on a Jolla. This was good as it is the reference platform for sailfish OS, but there has been a bug for the last several months which prevented me from updating the OS past version 1.1.7. Recently I got a Nexus 5, and was able to start testing Saera on it, only to discover that it is rather broken on 1.1.9. (Mainly due to GStreamer being updated.) I have started fixing it; currently I have got:

• the app launching without crashing;
• speech recognition working; and
• voice output working.

There are a few other things I need to fix (UI looks weird with different pixel densities, for example), but I hope to release the fix in the next few weeks. This should also fix issues anyone else may have with Saera on 1.1.8+.

Will it be available for other languages (not only English)?

I would like it to, however I am hampered by two things:

1. I would need translations for all the things Saera knows how to say (and someone who knows the language to help fix grammar-specific changes); and
2. I would need an acoustic model and a language model for Julius to be able to recognize speech.

The latter is more of a problem; I am currently using the VoxForge acoustic model. VoxForge is working on acoustic models in several languages (English, Bulgarian, Catalan, German, Greek, Spanish, French, and several others), but only English and German have a complete acoustic model. There is also a full Japanese model available from the Julius developers, as it is originally a Japanese project; I know very little about the Japanese language, however.

To summarize:

• German is possible; there is a language model and I know the language well enough to make the necessary changes.
• Japanese has a working language model, but I would need someone fluent in both Japanese and English with a bunch of free time to help me get Saera working there
• Other languages are not possible without either a) finding another source for acoustic voice models or b) using Internet voice recognition.

I'm interesting on Spanish language. How can I help you?

Find me voice recognition software for Spanish that runs on your device. Translate the sentences in intents.cfg into Spanish (square brackets are important words, angle brackets are variables that are expected to change). Make a spanish version of timeparser2.py, or find a spanish time parsing library, because Spanish formats times differently than English. Then of course we need translations for all the strings in saera2.py, and new API endpoints for weather, directions, business search etc that return directions in spanish.

In news: I've been working on the navigation system. The current one in openrepos is barely functional. I've been improving it to the point that I actually trust it for directions to somewhere I haven't been before. Notably, I changed the default routing provider from Graphhopper to Mapbox, which gives several improvements, such as saying "Go right at the fork" instead of "Continue" or "Take the exit ramp" instead of "Continue" (Graphhopper said "Continue" for everything that wasn't a turn and was therefore basically useless on highways).

Saera is difficult to understand when there is music playing too. I'm working on two solutions to this:

1. Lower the system volume and boost Saera's volume to 200%. This is based on what Android does to lower the volume of any other media while speaking. The problem is that this results in clipping and doesn't sound very good.
2. Pause music while Saera is talking and resume it after she finishes. This is better for comprehension and for audio quality, but may be slightly disorienting if something with a strong beat is briefly paused.

Thoughts on which option should be the default? Alternately, does anyone know how to change the volume for only one application instead of the whole system?