3

After spending quite a bit of time trying to design my way around the existing javax.sound.sampled library for simple (but relatively time-accurate) capture/rendering of audio, I'm concluding that I really need to get to the native audio API. The problem I've found is that I need to be able to get frame position of sound (that I'm rendering by writing to a SourceDataLine) in increments of 20 or 50 milliseconds, accurate to 1 or 2 milliseconds. In other words, I want to synchronize some graphical events to audio that is playing, with a video frame rate of 20 to 50 fps, with jitter no greater than 1 or 2 millseconds. It seems that the JavaSound implementation is too far from the hardware (a price to pay for platform independence?) for this. After much "tinkering" with SDL/TDL buffer sizes, audio formats, size of SDL.write() and TDL.read() buffers, frequency of writing, etc, I can't find a way to avoid the variation in the frame position reported compared to the system clock (at 1 millisecond resolution, not microseconds).

[Further clarification: I could just let the graphical rendering run at a fixed rate based on the system clock, which would be accurate enough for my needs. However, if this is to be synchronized with an audio track, by the time a typical audio track (5 minutes long, for instance) neared the end even small time differences will acccumulate to a noticeable time difference between audio and graphics. I thought to just check for synchronization every second (or 2 or 5), but the "jitter" from the reported audio frame position would result in corrections that would then be noticeable.]

I've been researching what libraries might already be available. It seems many of the existing packages are targeted to support the game development community. However, I don't need 3D audio, hardware acceleration of effects, synthesis, processing - just a way to read the frame position of what is being heard with a consistent latency. If the latency was as much as 100 milliseconds, but consistent, I could design around that.

I would appreciate recommendations for Java libraries providing this type of capability. I presume it will be JNI wrappers of a native audio API. I only need to support Windows7 at this point. I've seen OpenAL (seems to be Creative's "open" way of supporting hardware acceleration of audio since now no longer supported in Vista - game-oriented and more than I need), JSyn (focusing on synthesis, MIDI, not a simple sampled interface) and JAsioHost (most promising so far).

If there are ways around the limitations I've noted with the JavaSound API, that would also be fine, but I've just about given up after a week of effort.

Thanks.

Leif Gruenwoldt
  • 13,561
  • 5
  • 60
  • 64
ags
  • 719
  • 7
  • 23
  • For this kind of low latency audio, the only thing that can do this is ASIO, so you can go with JAsioHost, but you will need a soundcard with ASIO driver, which is mainly professional sound hardware. – Jakub Zaverka May 02 '12 at 21:30
  • @JakubZaverka Yes, I'm looking at ASIO. I have also found ASIO4ALL drivers (which work even for my tiny test laptop) but still this is more to complicate installation. I'm also seeing some strange behavior with the new drivers. I would prefer just simple frame position predictability/observability without this complication. I don't really need low latency (I do need to be able to measure a consistent latency to deal with it) but do need to be able to sync my application to the audio being played. – ags May 03 '12 at 00:31

2 Answers2

1

I haven't tried either of these yet, but may be worth looking into:

Leif Gruenwoldt
  • 13,561
  • 5
  • 60
  • 64
0

I guess there is no such library... but the task you want to accomplish is actually not that hard:

You need to synchronize your app to audio frames... Meaning the audio is the master authority in timing. You can create a new timer that will periodically post chunks of sound for playback and issue events that chunk xy is being played.

You need to supply chunks at the same rate as real time... meaning post one second worth of samples in one second.

The playback itself can be accomplished by WaveOut functions in Windows (link here, you can search for tutorials). You can access these functions from Java using JNI. The WaveOut plays the sound in Wav format, meaning you can open the device with simple LPC modulation and post raw samples.

The whole thing will not be totally synced up, but the delays would not be big. Watch out for buffer underflows for waveOutWrite, you need to supply new chunks BEFORE the last chunk finished playing, or the sound will be distorted.

Jakub Zaverka
  • 8,816
  • 3
  • 32
  • 48