WIP audio in server

[Jeremy Whiting]

Ok, another update. This time audio parameters are coming from the
user's config (I think, GlobalFDSet getting initialized is a mystery
to me so far because of macros and dotconf callbacks. Seems to work
here though.

BR,
Jeremy


On Mon, Aug 10, 2015 at 5:43 PM, Jeremy Whiting <jpwhiting at kde.org> wrote:
> Ok, here's a working patch. A few things I'll fix before this is ready
> for master though.
>
> 1. Audio initialization needs to come from the config files again.
> 2. Audio socket cleanup.
> 3. Documentation changes for this big change in how spd works.
> 4. How to request the server stop playing audio (or maybe it knows
> because it's telling the modules the same thing...
> 5. Audio file playback in generic.c needs to open the file and send
> the audio on the socket.
>
> BR,
> Jeremy
>
>
> On Wed, Aug 5, 2015 at 5:42 PM, Jeremy Whiting <jpwhiting at kde.org> wrote:
>> Ok, the audio in the server is in it's own thread now, and mostly all
>> code is in server/audio.c to keep it separate from the other file
>> descriptor handling for clients. I'm still getting pauses for some
>> reason, but it is threaded now at least and works when run with -d
>> also. I'll try to figure out where the pauses are coming from.
>>
>> BR,
>> Jeremy
>>
>> On Fri, Jul 31, 2015 at 6:47 PM, Jeremy Whiting <jpwhiting at kde.org> wrote:
>>> I've spent a bit of time wrestling with this code today and have found
>>> the following.
>>>
>>> If I don't initialize pulse by calling _pulse_open but only initialize
>>> once I have data to set the format it works mostly, though there are
>>> pauses of silence in long phrases from espeak still.
>>> If I do initialize pulse by calling _pulse_open in pulse_open and also
>>> reinitializing in pulse_play as required for audio rate changes etc.
>>> it doesn't play anything somehow (it's getting stuck in pa_simple_free
>>> on a mutex somehow).
>>>
>>> Do I need to run a thread for every audio socket we attach in the
>>> server? If so what should the thread's start_routine look like, just
>>> while(1) and the callback audio_process_incoming is called when we get
>>> traffic on the fd ? Once this is working (and I clean up how we
>>> initialize the audio to not hard coded values, but get the config from
>>> the config file) I can look at making the audio use the pulse async
>>> api, but I want to get the proof of concept working as a first step.
>>> My knowledge of pthreads seems to be blocking me at the moment though.
>>>
>>> thanks,
>>> Jeremy
>>>
>>>
>>>
>>> On Thu, Jul 30, 2015 at 7:42 PM, Jeremy Whiting <jpwhiting at kde.org> 
>>> wrote:
>>>> Oops, I didn't see this until just now. At any rate I got that issue
>>>> solved, now only playback failing (I verified I get samples in the
>>>> server and I think it's initializing the AudioId properly since it's
>>>> not NULL here. It seems to think it's playing from all the return
>>>> values, but I hear no audio yet. I also wasn't sure what to do about
>>>> sending the audio parameters just yet, so hard coded some based on my
>>>> config here for now.
>>>>
>>>> BR,
>>>> Jeremy
>>>>
>>>> On Thu, Jul 30, 2015 at 6:40 PM, Luke Yelavich
>>>> <luke.yelavich at canonical.com> wrote:
>>>>> On Fri, Jul 31, 2015 at 10:40:04AM AEST, Luke Yelavich wrote:
>>>>>> On Fri, Jul 31, 2015 at 07:27:27AM AEST, Jeremy Whiting wrote:
>>>>>> > Hey all,
>>>>>> >
>>>>>> > I'm implementing moving audio from the modules to the server (and
>>>>>> > modules will send audio data to the server on a unix socket). I've got
>>>>>> > the socket creation, and seem to have the ability to connect to the
>>>>>> > socket in the modules but it's hanging here when I try to run spd-say
>>>>>> > hello. Also I'm getting this in my speech-dispatcher.log as if it's
>>>>>> > trying to open a second audio connection from sd_espeak for some
>>>>>> > reason when it hangs (and no log output after this):
>>>>>> >
>>>>>> > [Thu Jul 30 15:03:15 2015 : 829380] speechd:     Adding audio
>>>>>> > connection on socket 4
>>>>>> > [Thu Jul 30 15:03:29 2015 : 105629] speechd:    Adding module on fd 28
>>>>>> > [Thu Jul 30 15:03:29 2015 : 105654] speechd:     Adding audio
>>>>>> > connection on socket 4
>>>>>> >
>>>>>> > I'm probably doing something obviously wrong, but can't seem to see
>>>>>> > what at the moment though I've been beating my head against it for a
>>>>>> > while and debugging. Can you see anything obvious in my changes?
>>>>>>
>>>>>> Well, I wonder if what you have in speechd_audio_connection_new is 
>>>>>> correct. You make reference to module sockets and the server socket 
>>>>>> where clients connect, and not the audio socket.
>>>>>
>>>>> Helps if I attach the diff.
>>>>>
>>>>> Luke
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-Moving-audio-from-modules-to-the-server.patch
Type: text/x-patch
Size: 57929 bytes
Desc: not available
URL: 
<http://lists.freebsoft.org/pipermail/speechd/attachments/20150810/8684ec1f/attachment-0001.bin>