DSP System Toolbox

Measuring Audio Latency

This example shows how to measure the latency introduced in audio signals when using either the dsp.AudioPlayer and dsp.AudioRecorder System objects in MATLAB or To Audio Device and From Audio Device blocks in Simulink. The procedure presented here allows user to tune certain parameters that affect the latency value. A list of these parameters is discussed later in the example.

Introduction

In broad terms, latency can be defined as the time it takes for the audio to be heard or recorded after it has been generated or played back, or vice versa. There are multiple factors that can influence this:

  1. Hardware being used: CPU, memory, sound card etc.

  2. Audio drivers that are used to communicate with the sound card.

  3. Algorithm complexity of audio processing between playback and recording.

  4. Software being used to process the audio: MATLAB or Simulink in this case.

This example shows you how to measure the overall latency, with focus on what parameters decide the latency effect of the last point in the above list. There are two flavors of this example: one uses the audio System objects in MATLAB and the other implements the same set-up in Simulink using the audio input and output blocks in DSP System Toolbox.

Experiment Set-up

In this example, we measure the latency by playing an audio signal through dsp.AudioPlayer (or To Audio Device block), recording the played audio through dsp.AudioRecorder (or From Audio Device block) and measuring the delay through cross-correlation of the two signals. No other processing is done on the audio signal. A loopback cable is used to physically connect the audio-out port of the sound card to its audio-in port. The set-up is shown in the illustration below:

Another common application set-up may be to record audio live via a microphone, process it on the computer and play it back in a streaming fashion. However, note that even in that case the hardware interface latency will still be the same as the above set-up: a sum of latencies due to the player and the recorder.

Measuring Latency

The function HelperMeasureAudioLatencyHelperMeasureAudioLatency computes the audio latency of the above setup. Inputs to this function are:

  • plotFlag: Set this to true if you want to visualize the input to dsp.AudioPlayer and the output from dsp.AudioRecorder objects.

  • frameSize: This is the number of samples of audio that constitute a single frame. This sets the SamplesPerFrame property of dsp.AudioFileReader object that reads the audio file to be played.

  • bufferSize: The value of this parameter is copied to the BufferSize property of dsp.AudioPlayer and dsp.AudioRecorder System objects.

  • playerQueueSize and recoderQueueSize: These are the queue size (in samples) for the dsp.AudioPlayer and dsp.AudioRecorder System objects, respectively. These are mapped to the QueueDuration property of these objects through the equation: QueueDuration = QueueSize/(Sample Rate).

  • playDuration: Number of seconds to play the audio.

  • useSimulink: If this is set to true, the Simulink model audiolatencymeasurementaudiolatencymeasurement is used to measure the latency.

Most of these inputs are the various parameters that affect audio latency in MATLAB or Simulink. These have been made as inputs so that the user can try different values and measure the latency for each set. The various sections in the HelperMeasureAudioLatency function are:

  1. Initialization: Default values for input parameters are assigned. Also, the audio System objects that will be used are created with appropriate parameter values.

  2. Loopback simulation: For duration specified by the playDuration input, audio is read from a file and sent to the computer's audio-out port through dsp.AudioPlayer. Through the loopback cable and the audio-in port, this audio is read in using dsp.AudioRecorder object. In the case when the useSimulink input is true, the loopback simulation is performed through the Simulink model audiolatencymeasurementaudiolatencymeasurement. The signal read from the file and the one receiver by the recorder after loopback are both stored in a signal sink.

  3. Cross-correlation: Using MATLAB's xcorr function, the cross-correlation between the audio signal sent to player and the one received by recorder is computed. The received signal will lag behind the played signal. The time shift that gives us maximum cross-correlation is decided as the latency of this set-up. Additionally, if the plotFlag input was true, the two audio signals are plotted in a figure so that the lag can be visually observed.

Parameters in MATLAB/Simulink That Affect Latency

Frame size

This isn't an explicit property of the player or recorder System objects, but rather the number of samples that are passed to those objects in each call to their step method. The bigger the frame size, the longer you'll have to wait to collect those many samples.

BufferSize

The BufferSize is a property of both dsp.AudioPlayer and dsp.AudioRecorder System objects. Its value will be half the size of the sound card buffer. The lower the BufferSize, the quicker you can send samples to the sound card. However, it is a trade-off because small buffer can also easily underrun (for player) or overrun (for recorder).

QueueDuration

Both dsp.AudioPlayer and dsp.AudioRecorder System objects have this property of QueueDuration. The queue is a storage space between the System object and the sound card buffer. Its main use is to help match the throughput of the algorithm in MATLAB and the device. In this regard, it can help prevent the above mentioned buffer underruns and overruns. Changing the QueueDuration usually has the most significant impact on latency. QueueDuration should be set to a value as low as possible to keep latency in check while not causing glitches in the audio.

The same parameters are also applicable in the Simulink version that uses audio blocks instead of System objects. More details about BufferSize and QueueDuration can be found in the documentation of the System objects dsp.AudioPlayerdsp.AudioPlayer and dsp.AudioRecorderdsp.AudioRecorder, and the blocks To Audio DeviceTo Audio Device and From Audio DeviceFrom Audio Device.

Sound Card and Drivers

Apart from the above three parameters, latency is also affected by the sound card and drivers. For low-latency applications, it is recommended to use a low-latency sound card. Drivers also have can have an impact on latency. For example, on Windows™ platform, DirectSound creates an additional virtual device layer between the application and the sound card, increasing latency. On the other hand, ASIO™ drivers talk to the device directly which helps to reduce delays.

Sample Results

Here are some results of running HelperMeasureAudioLatencyHelperMeasureAudioLatency on different input parameters:

latency = HelperMeasureAudioLatency(0, 4096, 256, 256, 256, 20, 0);

Result: Latency of 34ms

latency = HelperMeasureAudioLatency(1, 512, 512, 3072, 3072, 10, 1);

Result: Latency of 108ms

latency = HelperMeasureAudioLatency(1, 1024, 512, 2048, 2048, 20, 0);

Result: Latency of 64ms. You can also zoom into the plot to see the delay between player and recorder waveforms.

Summary

We discussed the concept of latency in audio signals and demonstrated a way in MATLAB and Simulink to quantify it through experiment. We also covered the various parameters that influence audio latency in MATLAB and Simulink. Some sample results of running that experiment on a computer were presented at the end. Note that the values used in examples above were particular to one machine. Results will vary from one machine to the next. It is recommended that you follow the guidance above and try different values to find the best combination that suits your system and application.