Portable C voice capture and (relatively) low latency feedback
https://github.com/type-one/CSideTone
Note: C++17 implementation available at https://github.com/type-one/SideTone Low latency implementation using C++17 and miniaudio duplex feature at https://github.com/type-one/SideToneDuplex
Small test program written in C11 to have voice feedback with reasonably short latency. The code is portable and use the miniaudio framework (https://miniaud.io/).
The purpose was to test a simple double buffer structure with frame locking, with a capture thread producing samples and a playing thread replaying what was captured. The double buffer itself is implemented in a cache-friendly way by splitting in two parts the internal contiguous buffer of an std::vector. The swapping of the buffers is also lock-free.
It was also fun to let my young children play with it and discover their own voice with a small delay producing an echo.
This small framework can be a start for more complex sound processing on the feedback voice.
Note: miniaudio has some full-duplex features that allow latency optimization on this scenario, but I didn't want to rely on that for the test.
Can be compiled and Linux and Windows, and should be easily adapted for other platforms supported by miniaudio (Mac, Android, BSD).
The underlying miniaudio uses the standard default sound API of the OS (Linux/ALSA, Windows/WinMM, Android/OpenSLES, Mac/CoreAudio...) and not a specific low latency one (Jack, ASIO, ...), with default parameters.
On Linux, just use cmake . On Windows, just use cmake-gui to generate a Visual Studio solution
Tested on Windows 10 and Ubuntu/Debian Linux (on native host and in VMplayer). It should work also on Raspberry Pi. It just use default capture and playback devices.
Launch in a shell using "csidetone.exe" or "csidetone" Press "enter" to exit
You can play with different parameters in the audio/audio_config.h file: format, sample rate, channels, period.
Recommended parameters are 22050 hz, mono (1 channel), signed 16, and a period between 10 and 40 ms. On specific audio boards and on native host, lower parameters can probably be used.
Compiling in release with O3/Ox/fast-math/SIMD optimisations can help to lower the period. Disabling the progress display can also help to lower the period.
Enjoy
Laurent Lardinois / Type One (TFL-TDV)