UWP AudioGraph API - FrameOutputNode 读取字节错误

Question

I have a problem with the FrameOutputNode of the UWP Audio Graph API. I have a very simple graph that reads audio from a wav (PCM 16000Hz, 16 bit mono) file and sends it to the frame output node for processing. When processing, I need the audio to be in shorts (as they are in the raw bytes of the file). But as I read here the data can only be read as floats.

Here is my code:

var encoding = MediaEncodingProfile.CreateWav(AudioEncodingQuality.Low);
encoding.Audio = AudioEncodingProperties.CreatePcm(16000, 1, 16);

AudioGraphSettings settings = new AudioGraphSettings(AudioRenderCategory.Media);
settings.EncodingProperties = encoding.Audio;
CreateAudioGraphResult result = await AudioGraph.CreateAsync(settings);
var graph = result.Graph;

var localFolder = Windows.Storage.ApplicationData.Current.LocalFolder;
StorageFile file = await localFolder.GetFileAsync("audio.wav");

var fileInputNodeResult = await graph.CreateFileInputNodeAsync(file);
var fileInputNode = fileInputNodeResult.FileInputNode;

fileInputNode.FileCompleted += async (AudioFileInputNode sender, object args) =>
{
    graph.Stop();
}

frameOutputNode = graph.CreateFrameOutputNode(encoding.Audio);
fileInputNode.AddOutgoingConnection(frameOutputNode);

graph.QuantumStarted+= AudioGraph_QuantumStarted;

With the following AudioGraph_QuantumStarted event handler:

private void AudioGraph_QuantumStarted(AudioGraph sender, object args)
{
    AudioFrame frame = frameOutputNode.GetFrame();
    ProcessFrameOutput(frame);
}

unsafe private void ProcessFrameOutput(AudioFrame frame)
{
    AudioBuffer buffer = frame.LockBuffer(AudioBufferAccessMode.Read);
    IMemoryBufferReference reference = buffer.CreateReference();
    ((IMemoryBufferByteAccess)reference).GetBuffer(out byte* dataInBytes, out uint capacityInBytes);

    if (capacityInBytes > 0) {
        // Read the first 20 bytes 
        for (int i = 0; i < 20; i++)
        {
            Debug.WriteLine(dataInBytes[i]);
        }
    }
}

The bytes I receive in the output are the following. Since the samples are returned as bytes of a float, I marked the sample boundary with a line.

0 0 0 0 | 0 0 0 184 | 0 0 128 184 | 0 0 0 184 ...

But when I read the actual bytes from the file with a byte reader:

FileStream fIn = new FileStream(@"/path/to/audio.wav", FileMode.Open);
BinaryReader br = new BinaryReader(fIn);
// Skip the first 44 bytes since they are header stuff
br.ReadBytes(44);
for (int i = 0; i < 20; i++) 
{
    Debug.WriteLine(br.ReadByte());
}

Then I get the actual bytes:

0 0 | 255 255 | 254 255 | 255 255 | 255 255 | 254 255 | 253 255 | 252 255 ...

Again I marked the individual samples (shorts -> two bytes) with a line.

As you can see the short bytes 255 255 somehow map to float bytes 0 0 0 184 as they reoccur. So what is that mapping? How can I retrieve the raw shorts from the floats? What do I need to do to actually read the wav file bytes?

Answer 1

My question was answered here . Basically, the floats are the range of the shorts -32768 to 32767 converted to range -1 to 1 in float.

So given a float x in the buffer (use (float*)dataInFloats = (float*)dataInBytes to convert) you can calculate the corresponding short with:

f(x) = (65535 * x - 1) / 2