I cannot convert between 16 bit Scaled PCM Data and Floating Point PCM Data in C++. I think I must be close because the output audio somewhat resembles what I expect, but it is distorted.
The reason I am doing this is because I am running ScummVM games in the browser. The ScummVM code runs on the server and my custom code posts audio and images up to a website. I am using the 'Web-Audio Api' to play sound in JavaScript on the front end.
I am trying to provide raw PCM data segmented by channel to the JavaScript. My logic here is that there will be less latency if no decoding is required.
I know that the audio data must be good, because I have successfully converted this into a wave file format and played it.
I am trying to get a multidimensional array of floats. The first dimension represents the channel and the second is the sample data for that channel.
I ask that we not discuss the usefulness of the project. This is a learning project for me, not an attempt to build a widely used product.
Here is my algorithm, with comments describing my reasoning.
C++ Side
typedef unsigned char byte;
double** OSystem_Cli::mixCallback(uint len)
{
const int NO_CHANNELS = 2;
double** result = new double*[NO_CHANNELS];
int sampleSize = len;
byte* samples = new byte[len];
_mixerImpl->mixCallback(samples, len); //Get the 16-bit PCM audio from Scumm VM. eg. [91,11,91,11 . . .] and stores it in samples
for (int channel = 0; channel < NO_CHANNELS; channel++)
{
for (int byteNo = channel * 2, channelByteNo = 0; byteNo < sampleSize - 1; byteNo = byteNo + NO_CHANNELS * 2, channelByteNo++)
{
if (channelByteNo == 0)
{
result[channel] = new double[sampleSize / NO_CHANNELS / 2];
}
unsigned short unsignedCombination = (static_cast<unsigned short>(samples[byteNo]) << 8) + samples[byteNo + 1]; //Join two bytes together to get 1 sixteen bit number.
short signedCombination;
memcpy(&signedCombination, &unsignedCombination, sizeof(signedCombination));
double signedFloat = static_cast<double>(signedCombination);
signedFloat = signedFloat / (float)32768; //Divide it to get the floating point representation, as https://stackoverflow.com/questions/15087668/how-to-convert-pcm-samples-in-byte-array-as-floating-point-numbers-in-the-range states.
if (signedFloat > 1)
{
signedFloat = 1;
}
else if (signedFloat < -1)
{
signedFloat = -1;
}
result[channel][channelByteNo] = signedFloat;
}
}
delete[] samples;
return result;
}
JavaScript Side (In type script):
pushOntoAudioQueue(pcmBytesByChannel: number[][]) {
const streamer: WebAudioStreamer = this;
const samplePartsPerChannel = pcmBytesByChannel[0].length;
const noChannels = pcmBytesByChannel.length;
if (this.audioStack.length > MaxQueueLength) {
this.audioStack = [];
}
const buffer = this.context.createBuffer(noChannels, samplePartsPerChannel, 16000); //Length is total number of bytes for all channels
for (let channel = 0; channel < noChannels; channel++) {
let float32ChannelData = new Float32Array(pcmBytesByChannel[channel].length);
float32ChannelData.set(pcmBytesByChannel[channel]);
buffer.copyToChannel(float32ChannelData, channel);
}
streamer.audioQueue.push(buffer);
}
