Converting 16-bit short to 32 bit float

Question

In the tone generator example for iOS:http://www.cocoawithlove.com/2010/10/ios-tone-generator-introduction-to.html

I am trying to convert a short array to Float32 in iOS.

        Float32 *buffer = (Float32 *)ioData->mBuffers[channel].mData;
        short* outputShortBuffer = static_cast<short*>(outputBuffer);

        for (UInt32 frame = 0, j=0; frame < inNumberFrames; frame++, j=j+2)
        {
            buffer[frame] =  outputShortBuffer[frame];
        }

For some reasons, I am hearing an added noise when played back from the speaker. I think that there is a problem with conversion from short to Float32?

score 3 · Accepted Answer · answered May 13 '13 at 20:53

Yes, there is.

Consider that the value-range for floating point samples is -1.0 <= Xn <= 1.0 and for signed short is -32767 <= Xn <= +32767. Merely casting will result in clipping on virtually all samples.

So taking this into account:

    Float32 *buffer = (Float32 *)ioData->mBuffers[channel].mData;
    short* outputShortBuffer = static_cast<short*>(outputBuffer);

    for (UInt32 frame = 0, j=0; frame < inNumberFrames; frame++, j=j+2)
    {
        buffer[frame] =  ((float) outputShortBuffer[frame]) / 32767.0f;
    }

[Note: this is not the optimal way of doing this].

However, are you sure your frames are mono? If not this might also be a cause of audio corruption as you'll only be copying one channel.

As an aside, why, if your output buffer is floats are you not using them throughout?

How to convert float array to short array? – yogesh Feb 21 '18 at 10:48 — yogesh, Feb 21 '18 at 10:48

score 0 · Answer 2 · answered Jun 05 '23 at 03:40

Here is a swift version.

    let floatBuffer = UnsafeMutableBufferPointer<Float32>(start: buffer.floatChannelData![0], count: frameLength)
    data.withUnsafeBytes { (bytes: UnsafeRawBufferPointer) in
        let int16Buffer = bytes.bindMemory(to: Int16.self)
        let float32Buffer = UnsafeMutableBufferPointer(start: floatBuffer.baseAddress, count: floatBuffer.count)
        (0..<frameLength).forEach { float32Buffer[$0] = Float32(int16Buffer[$0]) / 32767.0 }
    }

Converting 16-bit short to 32 bit float

2 Answers2