Sound
A few notes on sound, see also Category:Audio.
Contents
Software overview
Starting from user activities and working down...
Sound playback
You play back sound files in Browse. Vorbis and Speex are our preferred codecs for general audio and human speech. These compressed representations of audio are stored in files using the Ogg container file format. Ogg files generally have the extension .ogg.
The Totem browser plugin handles .ogg files. It uses the GStreamer multimedia framework to unpack Ogg containers and decode audio and video streams.
Sound generation and Music Activities
You create sounds in the TamTam suite of music and sound activities. Or you can try building your own sound/music activities using Csndsugui. OLPC activities use the Csound music and audio signal processing language as their audio "engine". Csound can play Ogg Vorbis files through specialised opcodes. However it is envisaged that the next version of libsndfile (which Csound uses) will allow for a more transparent and flexible IO to this file formats.
Sound recording
The Record activity also uses the GStreamer multimedia framework to record video and audio from the XO's camera and microphone.
Low-level
The XO's low-level sound API and kernel drivers are ALSA (wikipedia:Advanced Linux Sound Architecture). OSS programs can avail of the snd-pcm-oss module, which provides an emulation of /dev/dsp. This is currently not loaded by default, but can easily be done by the command
$ modprobe snd-pcm-oss
If this is working, then it's possible to test it using
$ cat /dev/dsp > /dev/dsp
Given that ALSA supports only single access to the audio device, it is important that activities/programs share it accordingly. This means releasing the device when out of focus, for instance. With the adoption of an audio server, this limitation disapppears.
We are looking into using PulseAudio sound server in the future, which has transparency across the network. Csound has an IO module to connect to it, which would allow a simpler migration path to this system.
Some Linux software packages require other sound software such as Jack (another audio server) and portaudio (an audio library used to access the audio device). Csound can use Jack and this is going to be tested, since this software is currently being included in the OS. Although Csound can also use portaudio, this is likely not to be used, as its direct Alsa support provides better performance.
Hardware
See Audio hardware
Speaker Capabilities
(most data from the leader of the TamTam project Ethrop who might have more data on the subject)
The speakers in the XO are from and for cellphone speakers. They are optimized for voice, and have less quality frequency response at the low end of the spectrum. The XO speakers have a severely biased frequency response. We have recently performed a thorough analysis of the audio response curve of the machine and there is a spectacular 12dB peak between 3000 and 4500 Hz, this on all models. I suspect these are mobile phone speakers designed for voice clarity. What this means is kids will likely crank up the volume so that they can hear some of the lower frequencies. Since the physical size of the speakers prohibits any frequencies below 350 HZ, as they try to get a decent bandwidth, they will get the "membrane-against-the-casing" distortion (which has the merit of making the kids lower the volume but risks killing the speakers if done routinely). Someone on the hardware side really should look at the long term prospects for audio hardware failure and see what correction we can bring, by limiting signal output and/or equalising the output of the AD1888 (we dont know what can be done on chip...)
The speakers start rolling off at about 600 Hz and are virtually worthless below 400 Hz.
The hardware has a one-pole highpass filter at about 400 Hz (I forget the exact frequency but it doesn't matter much) in order to reduce the amount of useless LF energy that is presented to the speakers. The rolloff is only in the speaker path; the headphone path has flat response across the audio band.
In my experience, equalization doesn't improve the sound from the speakers very much. They sound tinny and weak no matter what you do. Taming the big peak in the 4 Khz range is of some value, but most program material has little information in that region, so the perceived improvement is small. Boosting the bass makes things worse - the speakers don't have enough air-moving capacity (cone diameter times linear motion range) to render low frequencies, and sending them more signal just slams the mechanical structure against its physical limits, causing distortion and possible damage.
In listening to podcasts, certainly headphones sound better.
Random bits
DTMF
For your collective interest, the speakers can reproduce DTMF tones reliably provided the levels are set down from maximum.
At lunch today on a B2 with build-debian, the dtmfdial package was used to transmit tones over a ham radio for making an IRLP request. The DTMF tones include 697 Hz for the top row.
Overcoming lack of Tonic
Music activities should thus default to a bassoon.
The odd thing about a bassoon is that the fundamental frequency is nearly absent. The ear-brain system fills in this frequency, making the bassoon sound very low pitched without actually containing much of the very low frequencies.
At the other extreme, a sine wave is worst case. Recorders produce this, and flutes nearly do.
