Version 0.10 includes major API changes and refactoring, mostly concerned with the data conversion system. Data conversion is performed internally to convert audio data between the format requested when initializing the ma_device object and the format of the internal device used by the backend. The same applies to the ma_decoder object. The previous design has several design flaws and missing features which necessitated a complete redesign.
Changes to Data Conversion
The previous data conversion system used callbacks to deliver input data for conversion. This design works well in some specific situations, but in other situations it has some major readability and maintenance issues. The decision was made to replace this with a more iterative approach where you just pass in a pointer to the input data directly rather than dealing with a callback.
The following are the data conversion APIs that have been removed and their replacements:
- ma_format_converter -> ma_convert_pcm_frames_format()
- ma_channel_router -> ma_channel_converter
- ma_src -> ma_resampler
- ma_pcm_converter -> ma_data_converter
The previous conversion APIs accepted a callback in their configs. There are no longer any callbacks to deal with. Instead you just pass the data into the *_process_pcm_frames() function as a pointer to a buffer.
The simplest aspect of data conversion is sample format conversion. To convert between two formats, just call ma_convert_pcm_frames_format(). Channel conversion is also simple which you can do with ma_channel_converter via ma_channel_converter_process_pcm_frames().
Resampling is more complicated because the number of output frames that are processed is different to the number of input frames that are consumed. When you call ma_resampler_process_pcm_frames() you need to pass in the number of input frames available for processing and the number of output frames you want to output. Upon returning they will receive the number of input frames that were consumed and the number of output frames that were generated.
The ma_data_converter API is a wrapper around format, channel and sample rate conversion and handles all of the data conversion you'll need which probably makes it the best option if you need to do data conversion.
In addition to changes to the API design, a few other changes have been made to the data conversion pipeline:
- The sinc resampler has been removed. This was completely broken and never actually worked properly.
- The linear resampler now uses low-pass filtering to remove aliasing. The quality of the low-pass filter can be controlled via the resampler config with the lpfOrder option, which has a maximum value of MA_MAX_FILTER_ORDER.
- Data conversion now supports s16 natively which runs through a fixed point pipeline. Previously everything needed to be converted to floating point before processing, whereas now both s16 and f32 are natively supported. Other formats still require conversion to either s16 or f32 prior to processing, however ma_data_converter will handle this for you.
Custom Memory Allocators
miniaudio has always supported macro level customization for memory allocation via MA_MALLOC, MA_REALLOC and MA_FREE, however some scenarios require more flexibility by allowing a user data pointer to be passed to the custom allocation routines. Support for this has been added to version 0.10 via the ma_allocation_callbacks structure. Anything making use of heap allocations has been updated to accept this new structure.
The ma_context_config structure has been updated with a new member called allocationCallbacks. Leaving this set to it's defaults returned by ma_context_config_init() will cause it to use MA_MALLOC, MA_REALLOC and MA_FREE. Likewise, The ma_decoder_config structure has been updated in the same way, and leaving everything as-is after ma_decoder_config_init() will cause it to use the same defaults.
The following APIs have been updated to take a pointer to a ma_allocation_callbacks object. Setting this parameter to NULL will cause it to use defaults. Otherwise they will use the relevant callback in the structure.
- ma_rb_init() / ma_rb_init_ex()
- ma_pcm_rb_init() / ma_pcm_rb_init_ex()
Note that you can continue to use MA_MALLOC, MA_REALLOC and MA_FREE as per normal. These will continue to be used by default if you do not specify custom
Buffer and Period Configuration Changes
The way in which the size of the internal buffer and periods are specified in the device configuration have changed. In previous versions, the config variables bufferSizeInFrames and bufferSizeInMilliseconds defined the size of the entire buffer, with the size of a period being the size of this variable divided by the period count. This became confusing because people would expect the value of bufferSizeInFrames or bufferSizeInMilliseconds to independantly determine latency, when in fact it was that value divided by the period count that determined it. These variables have been removed and replaced with new ones called periodSizeInFrames and periodSizeInMilliseconds.
These new configuration variables work in the same way as their predecessors in that if one is set to 0, the other will be used, but the main difference is that you now set these to you desired latency rather than the size of the entire buffer. The benefit of this is that it's much easier and less confusing to configure latency.
The following unused APIs have been removed:
The following macros have been removed:
Other API Changes
Other less major API changes have also been made in version 0.10.
ma_device_set_stop_callback() has been removed. If you require a stop callback, you must now set it via the device config just like the data callback.
The ma_sine_wave API has been replaced with a more general API called ma_waveform. This supports generation of different types of waveforms, including sine, square, triangle and sawtooth. Use ma_waveform_init() in place of ma_sine_wave_init() to initialize the waveform object. This takes a configuration object called ma_waveform_config which defines the properties of the waveform. Use ma_waveform_config_init() to initialize a ma_waveform_config object. Use ma_waveform_read_pcm_frames() in place of ma_sine_wave_read_f32() and ma_sine_wave_read_f32_ex().
ma_convert_frames() and ma_convert_frames_ex() have been changed. Both of these functions now take a new parameter called frameCountOut which specifies the size of the output buffer in PCM frames. This has been added for safety. In addition to this, the parameters for ma_convert_frames_ex() have changed to take a pointer to a ma_data_converter_config object to specify the input and output formats to convert between. This was done to make it more flexible, to prevent the parameter list getting too long, and to prevent API breakage whenever a new conversion property is added.
ma_calculate_frame_count_after_src() has been renamed to ma_calculate_frame_count_after_resampling() for consistency with the new ma_resampler API.
The following filters have been added:
- ma_biquad - Biquad filter (transposed direct form 2)
- ma_lpf1 - First order low-pass filter
- ma_lpf2 - Second order low-pass filter
- ma_lpf - High order low-pass filter (Butterworth)
- ma_hpf1 - First order high-pass filter
- ma_hpf2 - Second order high-pass filter
- ma_hpf - High order high-pass filter (Butterworth)
- ma_bpf2 - Second order band-pass filter
- ma_bpf - High order band-pass filter
- ma_peak2 - Second order peaking filter
- ma_notch2 - Second order notching filter
- ma_loshelf2 - Second order low shelf filter
- ma_hishelf2 - Second order high shelf filter
These filters all support 32-bit floating point and 16-bit signed integer formats natively. Other formats need to be converted beforehand.
Sine, Square, Triangle and Sawtooth Waveforms
Previously miniaudio supported only sine wave generation. This has now been generalized to support sine, square, triangle and sawtooth waveforms. The old ma_sine_wave API has been removed and replaced with the ma_waveform API. Use ma_waveform_config_init() to initialize a config object, and then pass it into ma_waveform_init(). Then use ma_waveform_read_pcm_frames() to read PCM data.
A noise generation API has been added. This is used via the ma_noise API. Currently white, pink and Brownian noise is supported. The ma_noise API is similar to the waveform API. Use ma_noise_config_init() to initialize a config object, and then pass it into ma_noise_init() to initialize a ma_noise object. Then use ma_noise_read_pcm_frames() to read PCM data.
Internal functions have all been made static where possible. If you get warnings about unused functions, please submit a bug report.
The ma_device structure is no longer defined as being aligned to MA_SIMD_ALIGNMENT. This resulted in a possible crash when allocating a ma_device object on the heap, but not aligning it to MA_SIMD_ALIGNMENT. This crash would happen due to the compiler seeing the alignment specified on the structure and assuming it was always aligned as such and thinking it was safe to emit alignment-dependant SIMD instructions. Since miniaudio's philosophy is for things to just work, this has been removed from all structures.
Results codes have been overhauled. Unnecessary result codes have been removed, and some have been renumbered for organisation purposes. If you are are binding maintainer you will need to update your result codes. Support has also been added for retrieving a human readable description of a given result code via the ma_result_description() API.