Skip to content

Writing A Custom Microphone Capture System

By default Dissonance uses the BasicMicrophoneCapture behaviour to record audio from a microphone using the Unity Microphone API and feed it into Dissonance. However this script is not ideal for all use cases. You can replace the microphone capture system in Dissonance by creating a new behaviour which implements IMicrophoneCapture and adding the script to the same gameObject as the DissonanceComms behaviour.

This tutorial will explain how to build a replacement capture system which streams audio from a file. before following the tutorial make sure you've read the reference docs so you understand what the IMicrophoneCapture interface means.

Step 1 : Basic Setup⚓︎

First you need to create a new script with the IMicrophoneCapture interface on it and drop it onto the same GameObject as the DissonanceComms component.

Here is an example script. If you run the scene with this you should see a single exception printed to the console coming from the StartCapture method.

Step 2 : Start And Stop⚓︎

Now you need to properly start and stop the script without throwing exceptions.

StartCapture should return the format of the audio you will be providing. This must be mono (i.e. 1 channel) and any sample rate is acceptable (just use whatever is most convenient for you). If your capture system is not ready you can return null to prevent start-up. If you return a non-null value you must set IsRecording to true and you should set Latency to an appropriate value. The Latency value indicates an estimate of the time between sound physically hitting the microphone to submitting the audio to Dissonance, if you don't know this value leave it set to zero.

StopCapture should do whatever you need to stop the underlying capture system. Once this is done you must set IsRecording to false.

Subscribe and Unsubscribe should simply keep a list of subscribers. You can implement this as a List<IMicrophoneSubscriber> where Subscribe just calls Add and Unsubscribe just calls Remove and returns the value.

here is an example script. If you run the scene with this you should see exceptions printed to the console every frame coming from the UpdateSubscribers method.

Step 3 : Streaming Silence⚓︎

Now you need to stream some audio to Dissonance to stop the script throwing exceptions every frame.

When IsRecording is true (i.e. after StartCapture has been called and before StopCapture has been called) your capture script must provide audio at approximately a realtime rate. Dissonance will try to handle slight "bumps" (e.g. audio arriving slightly early or late) but overall you must supply audio at the correct rate. For example the BasicMicrophoneCapture script assumes that the microphone supplies audio at the correct rate, if you're reading from some kind of recording hardware this is probably a good assumption to make. The basic process for the microphone capture (which you may be able to replicate in your custom system) is:

  1. Drain all data from the recording hardware into a buffer
  2. Copy as much data out of the buffer into a preallocated array as possible
  3. Submit preallocated array to subscribers
  4. if there is data left in the buffer, goto 2

For this step we won't interact with any hardware, instead we'll just submit silence to Dissonance at the correct rate.

here is an example script which implements UpdateSubscribers by simply submitting silence at the correct rate. If you run this everything should work as expected in Dissonance (no exceptions), but of course you will not hear anything. This works by preallocating an array of 960 float, which represents 20ms of audio at 48kHz sample rate. Every time 20ms have elapsed, the buffer is submitted to Dissonance. Note that time is measured using unscaledDeltaTime, since audio needs to run at real time rate.

Step 4 : File Streaming⚓︎

Finally we'll add a basic file streaming system, this will read an audio file and play it into Dissonance. For simplicity this will not handle decoding of the audio from any well know format, instead you should transcode the audio into raw samples. You can do this with ffmpeg:

ffmpeg.exe -re -i AudioFile.wav -f f32le -ar 48000 -ac 1 output.raw

here is an example script which implements this. The basic process is the same as the silence system:

  1. Count time passed using unscaledDeltaTime
  2. When 20ms have passed, read up to 20ms of audio from the file (as bytes)
  3. Copy bytes into float buffer
  4. Submit to all subscribers