Audio Buffering in Android

Question

I'm trying to implement efficient audio buffer for radio streaming. Initally I was trying to do buffering as follows

So initially I have my InputStream which is taken from HttpUrlConnection. The data is being read from input stream by circular buffer which fills up the buffers. In this case I have 3 buffers but the amount and size of buffers can be easily changed. So after all buffers are filled up I start to write data from my first buffer to the OutputStream which is connected to InputStream, so whenever I write something to the OutputStream it can be then read from InputStream. MediaCodec is responsible for reading the data from InputStream and decoding the data which is then being passed to AudioTrack.

The problem with this set up is that after a while OutputStream eventually reaches the end of circular buffer, so there is no more "extra buffer" in between "OutputStream.giveMeNextBufferWithData" and "CircularBuffer.readDataFromInputStreamAndCreateBuffer" (pseudo code).

I've tried increasing the amount and size of each buffer, but that didn't help. It just delays the amount of time before OutputStream gets "hungry" for more data.

There is one very good library https://code.google.com/p/aacdecoder-android/ which works very well when it comes to buffer audio, but unfortunately I couldn't figure out how the buffering is done there and it doesn't work that well under some circumstances.

So is there any "perfect" algorithm for such a task ? I was thinking about double/triple buffering but I'm not too sure about it and considering that I've been searching around the net and implementing different solutions with no success for last few days I eventually decided to ask her.

I hope I explained everything well.

Thank you all in advance !

Answer 1

Buffering, by itself, is ultimately insufficient for fixed rate broadcast (vs on-demand) stream sources, where you cannot guarantee that the source and device sample rates are exactly the same.

For short programs, what you can do is buffer a second or two of audio, and if your playback rate is fast you will start to dip into that reserve, but hopefully not exceed it. Similarly, you need excess buffer capacity in case you are playing back data more slowly than it is generated. But these will ultimately fail when the sample rate difference accumulates to more than the buffer reserve size.

Instead, for streaming to work "eternally" you need a way to equalize the production and consumption rates.

A "lazy" way to do this is to have logic which adds or drops one sample per some longer unit of time, depending on the direction the error has been trending. But that technically introduces distortion.

The "fancy" way to do it is to have a variable sample rate converter the ratio of which can be adjusted over time to counteract the error between the audio consumption rate and the production rate.