python-sounddevice: A Python repository from wrongbad

2018 Jun 18 - Created fast-path hooks for numba-compiled callbacks to enable best possible performance of audio callbacks written in python.

2018 Jun 24 - Decided it's non trivial to use jitclasses for this. Realized even numpy's frombuffer allocates new memory, as such is very weak for stable real-time loops. E.g. alt-tabbing causes audio glitches, presumably due to contention for malloc.

2018 Jun 30 - Discovered cling and cling jupyter kernel. Covers main reasons I wanted to use python (OOP, interactive, in-context audio dev) can explicitly avoid malloc, can do 1-sample callbacks with no window manager inducable gliches. Abandoning this fork.

Play and Record Sound with Python

This Python module provides bindings for the PortAudio library and a few convenience functions to play and record NumPy arrays containing audio signals.

Documentation:: http://python-sounddevice.readthedocs.io/
Source code repository and issue tracker:: https://github.com/spatialaudio/python-sounddevice/
License:: MIT -- see the file LICENSE for details.

Requirements

Python:

Of course, you'll need Python. Any version where CFFI (see below) is supported should work. If you don't have Python installed yet, you should get one of the distributions which already include CFFI and NumPy (and many other useful things), e.g. Anaconda or WinPython.

pip/setuptools:

Those are needed for the installation of the Python module and its dependencies. Most systems will have these installed already, but if not, you should install it with your package manager or you can download and install pip and setuptools as described on the pip installation page. If you happen to have pip but not setuptools, use this command:

python3 -m pip install setuptools --user

To upgrade to a newer version of an already installed package (including pip itself), use the --upgrade flag.

CFFI:

The C Foreign Function Interface for Python is used to access the C-API of the PortAudio library from within Python. It supports CPython 2.6, 2.7, 3.x; and is distributed with PyPy. If it's not installed already, you should install it with your package manager (the package might be called python3-cffi or similar), or you can get it with:

python3 -m pip install cffi --user

PortAudio library:

The PortAudio library must be installed on your system (and CFFI must be able to find it). Again, you should use your package manager to install it (the package might be called libportaudio2 or similar). If you prefer, you can of course also download the sources and compile the library yourself. If you are using Mac OS X or Windows, the library will be installed automagically with pip (see "Installation" below).

NumPy (optional):

NumPy is only needed if you want to play back and record NumPy arrays. The classes sounddevice.RawStream, sounddevice.RawInputStream and sounddevice.RawOutputStream use plain Python buffer objects and don't need NumPy at all. If you need NumPy, you should install it with your package manager or use a Python distribution that already includes NumPy (see above). You can also install NumPy with pip, but depending on your platform, this might require a compiler and several additional libraries:

python3 -m pip install NumPy --user

Installation

Once you have installed the above-mentioned dependencies, you can use pip to download and install the latest release with a single command:

python3 -m pip install sounddevice --user

If you want to install it system-wide for all users (assuming you have the necessary rights), you can just drop the --user option. If you have installed the module already, you can use the --upgrade flag to get the newest release.

To un-install, use:

python3 -m pip uninstall sounddevice

If you are using Windows, you can alternatively install one of the packages provided at https://www.lfd.uci.edu/~gohlke/pythonlibs/#sounddevice. The PortAudio library is also included in the package and you can get the rest of the dependencies on the same page.

Usage

First, import the module:

import sounddevice as sd

Playback

Assuming you have a NumPy array named myarray holding audio data with a sampling frequency of fs (in the most cases this will be 44100 or 48000 frames per second), you can play it back with sounddevice.play():

sd.play(myarray, fs)

This function returns immediately but continues playing the audio signal in the background. You can stop playback with sounddevice.stop():

sd.stop()

If you know that you will use the same sampling frequency for a while, you can set it as default using sounddevice.default.samplerate:

sd.default.samplerate = fs

After that, you can drop the samplerate argument:

sd.play(myarray)

Recording

To record audio data from your sound device into a NumPy array, use sounddevice.rec():

duration = 10.5  # seconds
myrecording = sd.rec(int(duration * fs), samplerate=fs, channels=2)

Again, for repeated use you can set defaults using sounddevice.default:

sd.default.samplerate = fs
sd.default.channels = 2

After that, you can drop the additional arguments:

myrecording = sd.rec(duration * fs)

This function also returns immediately but continues recording in the background. In the meantime, you can run other commands. If you want to check if the recording is finished, you should use sounddevice.wait():

sd.wait()

If the recording was already finished, this returns immediately; if not, it waits and returns as soon as the recording is finished.

Alternatively, you could have used the blocking argument in the first place:

myrecording = sd.rec(duration * fs, blocking=True)

By default, the recorded array has the data type 'float32' (see sounddevice.default.dtype), but this can be changed with the dtype argument:

myrecording = sd.rec(duration * fs, dtype='float64')

Simultaneous Playback and Recording

To play back an array and record at the same time, use sounddevice.playrec():

myrecording = sd.playrec(myarray, fs, channels=2)

The number of output channels is obtained from myarray, but the number of input channels still has to be specified.

Again, default values can be used:

sd.default.samplerate = fs
sd.default.channels = 2
myrecording = sd.playrec(myarray)

In this case the number of output channels is still taken from myarray (which may or may not have 2 channels), but the number of input channels is taken from sounddevice.default.channels.

Device Selection

In many cases, the default input/output device(s) will be the one(s) you want, but it is of course possible to choose a different device. Use sounddevice.query_devices() to get a list of supported devices. The same list can be obtained from a terminal by typing the command

python3 -m sounddevice

You can use the corresponding device ID to select a desired device by assigning to sounddevice.default.device or by passing it as device argument to sounddevice.play(), sounddevice.Stream() etc.

Instead of the numerical device ID, you can also use a space-separated list of case-insensitive substrings of the device name (and the host API name, if needed). See sounddevice.default.device for details.

import sounddevice as sd
sd.default.samplerate = 44100
sd.default.device = 'digital output'
sd.play(myarray)

Callback Streams

Callback "wire" with sounddevice.Stream:

import sounddevice as sd
duration = 5.5  # seconds

def callback(indata, outdata, frames, time, status):
    if status:
        print(status)
    outdata[:] = indata

with sd.Stream(channels=2, callback=callback):
    sd.sleep(int(duration * 1000))

Same thing with sounddevice.RawStream:

import sounddevice as sd
duration = 5.5  # seconds

def callback(indata, outdata, frames, time, status):
    if status:
        print(status)
    outdata[:] = indata

with sd.RawStream(channels=2, dtype='int24', callback=callback):
    sd.sleep(int(duration * 1000))

Note

We are using 24-bit samples here for no particular reason (just because we can).

Blocking Read/Write Streams

Instead of using a callback function, you can also use the blocking methods sounddevice.Stream.read() and sounddevice.Stream.write() (and of course the corresponding methods in sounddevice.InputStream, sounddevice.OutputStream, sounddevice.RawStream, sounddevice.RawInputStream and sounddevice.RawOutputStream).

wrongbad/python-sounddevice