This is a bindings library to libmixed, an audio mixing and processing library.
Precompiled versions of the underlying library are included in this. If you want to build it manually however, refer to the libmixed repository.
First, set up your input and output parts such as file decoders and audio systems. Handling this is not a part of libmixed. To see an example of how to incorporate them, see the test.lisp file.
Load the system through ASDF or Quicklisp:
Now you'll need to integrate your inputs and outputs with the Mixed system. In order to do that, you'll want an unpacker and a packer segment. Both of those create "packed-audio", which holds all the information about how the raw audio data is encoded in a byte buffer.
(cl-mixed:make-unpacker c-buffer bytes sample-encoding channel-count channel-layout samplerate)
If you don't already have a byte buffer from your input or output implementation, passing NIL for the
c-buffer will automatically create one for you, which you can then access with
data. An example call might look like this:
(cl-mixed:make-unpacker NIL 4096 :int16 2 :alternating 44100)
An optional third parameter designates the sample rate of the buffers that the source converts to or the drain converts from. This "buffer sample rate" has to be the same across all segments in a mixer pipeline. It defaults to 44100. Creating a packer looks and works exactly the same as an unpacker.
Next you'll want to create the segments that'll do the actual audio processing you want. For this example, let's create a 3D audio segment (
space-mixer) and a fade effect (
(cl-mixed:make-space-mixer) (cl-mixed:make-fader :duration 5.0)
Next we'll need to create the buffers that are used to manipulate the audio internally and bind them to the appropriate inputs and outputs of our segments.
(cl-mixed:with-buffers 500 (input left right) (cl-mixed:connect unpacker :left fader :mono input) (setf (cl-mixed:output :right unpacker) right) (cl-mixed:connect fader :mono space-mixer 0 input) (cl-mixed:connect space-mixer :left packer :left left) (cl-mixed:connect space-mixer :right packer :right right) ...)
This here means we create three buffers,
right, each with a size capable of holding 500 samples. We then connect the source's left output to the fader's single input. Then we connect the right buffer to the unpacker's right output, just so that it has both outputs set. If your unpacker only has one channel, you can leave that out. If it has more, you'll have to repeat it for the other channels as well. Next we connect the fader's output as the space-mixer's first input. Finally we connect the left and right outputs of the space-mixer segment to the left and right inputs of the packer respectively. For the fader segment we can connect the same buffer to both input and output, as it is declared to work "in place". For the space-mixer segment we need distinct buffers, hence the extra
Now we can create our segment-sequence object, which keeps the order in which to process each segment.
(cl-mixed:make-segment-sequence source fader space drain)
Finally we can move to the main processing loop, which should look as follows:
(cl-mixed:start segment-sequence) (unwind-protect (loop while has-more do (process-source) (cl-mixed:mix 500 segment-sequence) (process-drain)) (cl-mixed:end segment-sequence))
process-drain are functions that will cause your source to put samples into its buffer and drain to read out the samples from its buffer. Running this now will just give you a fade in effect, which isn't too exciting. Since we haven't actually set or changed any of the 3D audio parameters, that effect remains inaudible. Changing the loop body to read something like the following
for tt = 0 then (+ tt 0.001) for dx = 0 then (- (* 100 (sin tt)) x) for dz = 0 then (- (* 50 (cos tt)) z) for x = (* 100 (sin tt)) then (+ x dx) for z = (* 50 (cos tt)) then (+ z dz) do (setf (cl-mixed:input-field :location 0 space-mixer) (list x 0 z)) (setf (cl-mixed:input-field :velocity 0 space-mixer) (list dx 0 dz)) (process-source) (cl-mixed:mix 500 segment-sequence) (process-drain)
Should cause the source to now also circle around the listener as it is fading in. If you change the
tt change factor from
0.001 to something higher, it will circle faster, and the doppler effect should become more noticeable.
The following segments are included with the standard libmixed distribution:
basic-mixerLinearly mix multiple inputs.
delayDelay the input by a given amount of time.
faderFade the volume of a source in or out.
frequency-passFilter out low or high frequencies.
generatorGenerate simple wave forms.
ladspaUse a LADSPA plugin.
pitchShift the pitch.
queueQueue multiple segments one after the other.
repeatRecord an input and then repeat it back.
space-mixerMix multiple inputs as if they were in 3D space.
volume-controlAdapt the volume and pan of a stereo signal.
See the next section on how to make custom segments.
Creating Custom Segments
While it is perfectly possible to create custom segments in C and load them into your lisp image, you can also write them directly in CL. This may be desired if performance is not crucial or if you want to quickly prototype an effect before translating it to a lower-level language.
In order to create a segment in Lisp, you must subclass
virtual and in the very least implement the
mix method for it. Here's an example for a very primitive echo effect:
(defclass echo (cl-mixed:virtual) ((buffer :initform NIL :accessor buffer) (offset :initform 0 :accessor offset) (delay :initarg :delay :initform 0.2 :accessor delay) (falloff :initarg :falloff :initform 0.8 :accessor falloff) (samplerate :initarg :samplerate :initform 44100 :accessor samplerate))) (defmethod cl-mixed:start ((echo echo)) (setf (buffer echo) (make-array (ceiling (* (delay echo) (samplerate echo))) :element-type 'single-float :initial-element 0.0f0))) (defmethod cl-mixed:mix (samples (echo echo)) (let ((out (cl-mixed:data (aref (cl-mixed:outputs echo) 0))) (in (cl-mixed:data (aref (cl-mixed:inputs echo) 0))) (buf (buffer echo)) (offset (offset echo)) (falloff (falloff echo))) (loop for i from 0 below samples for sample = (cffi:mem-aref in :float i) for echo = (aref buf offset) do (setf (cffi:mem-aref out :float i) (+ sample echo)) (setf (aref buf offset) (* (+ sample echo) falloff)) (setf offset (mod (1+ offset) (length buf)))) (setf (offset echo) offset) T))
In order to achieve the echo effect we keep samples of a given duration around in a ring buffer and then decrease their potency with each iteration while adding the new samples on top. Of course, a more natural sounding echo effect would need more complicated processing than this. Regardless, this segment can now be integrated just the same as the
fader segment from the above introductory code. The
T at the end of the
mix method signals that we're still producing samples. For segments that can end, the return value should become
NIL once they're exhausted.
A buffer holds an array of float-encoded audio samples. Everything within libmixed deals in float samples and reads from these buffers and/or writes to these buffers. This means everything that processes audio will be able to work within the same constraints of 32-bit float encoded samples, without having to worry about different sample rates, sample encodings, or channel layouts. Buffers have a fixed number of samples that they can hold, which should typically be consistent throughout the entire system.
The packed-audio is a representation of audio data that is packed into a single array and isn't in the standard buffer format. Many libraries that decode or encode audio files, play audio back, or process audio in some other way expect a single audio sample array of a particular samplerate, sample encoding, and channel layout, rather than the standardised sample buffers that libmixed uses. The packed-audio allows you to handle the conversion from this single array, encoded format, to the buffer format of libmixed and back.
Particularly relevant are the
packer segments, which you can use to handle the edges of the pipeline where other libraries interact with libmixed.
Libmixed allows you to define a pipeline to process audio in. This pipeline -- or graph, if you will -- is pieced together by segments that exchange data between each other through buffers. Each segment has a number of inputs, outputs, and fields that you can set and get. Every input and output can also have a differing amount of applicable fields, but each must have at least the buffer field, which designates the buffer that is connected at that point. You can retrieve information about how many inputs and outputs the segment expects or supports as well as information about the fields it understands by using the
info function on a segment.
Aside from the inputs, outputs, and fields, each segment has three methods that are central to the mixing of audio:
end allow you to prepare and clean up work shortly before and after mixing has been done. This can be important for real-time audio processing that cannot afford long pauses. The
mix method performs the actual mixing operation and should cause new samples to appear in the outputs' buffers.
Mixers are segments that take a run-time variable number of inputs and mix them together to a single output per channel. Libmixed provides two standard mixers out of the box,
space-mixer, which should serve most needs.
A segment-sequence simply ties together a number of segments and performs the
end operations in the order the segments were added to the sequence. This is mostly for convenience, in order to quickly perform the actual mixing, once the pipeline has been completely assembled already.
- Nicolas Hafner <email@example.com>
- Nicolas Hafner <firstname.lastname@example.org>