SoundCard

SoundCard is a library for playing and recording audio without resorting to a CPython extension. Instead, it is implemented using the wonderful CFFI and the native audio libraries of Linux, Windows and macOS.

SoundCard is cross-platform, and supports Linux/pulseaudio, Mac/coreaudio, and Windows/WASAPI. While the programming interface is identical across platforms, sound card naming schemes and default block sizes can vary between devices and platforms.

SoundCard is still in development. All major features work on all platforms, but there are a few known issues that still need to be fixed. If you find a bug, please open an Issue, and I will try to fix it. Or open a Pull Request, and I will try to include your fix into SoundCard.

However, please be aware that this is a hobby project of mine that I am developing for free, and in my spare time. While I try to be as accomodating as possible, I can not guarantee a timely response to issues. Publishing Open Source Software on Github does not imply an obligation to fix your problem right now. Please be civil.

SoundCard is licensed under the terms of the BSD 3-clause license

(c) 2016 Bastian Bechtold

Tutorial

Here is how you get to your Speakers and Microphones:

import soundcard as sc

# get a list of all speakers:
speakers = sc.all_speakers()
# get the current default speaker on your system:
default_speaker = sc.default_speaker()
# get a list of all microphones:
mics = sc.all_microphones()
# get the current default microphone on your system:
default_mic = sc.default_microphone()

# search for a sound card by substring:
>>> sc.get_speaker('Scarlett')
<Speaker Focusrite Scarlett 2i2 (2 channels)>
>>> one_mic = sc.get_microphone('Scarlett')
<Microphone Focusrite Scalett 2i2 (2 channels)>
# fuzzy-search to get the same results:
one_speaker = sc.get_speaker('FS2i2')
one_mic = sc.get_microphone('FS2i2')

All of these functions return Speaker and Microphone objects, which can be used for playback and recording. All data passed in and out of these objects are frames × channels Numpy arrays.

import numpy

>>> print(default_speaker)
<Speaker Focusrite Scarlett 2i2 (2 channels)>
>>> print(default_mic)
<Microphone Focusrite Scarlett 2i2 (2 channels)>

# record and play back one second of audio:
data = default_mic.record(samplerate=48000, numframes=48000)
default_speaker.play(data/numpy.max(data), samplerate=48000)

# alternatively, get a `Recorder` and `Player` object
# and play or record continuously:
with default_mic.recorder(samplerate=48000) as mic, \
      default_speaker.player(samplerate=48000) as sp:
    for _ in range(100):
        data = mic.record(numframes=1024)
        sp.play(data)

Latency

By default, SoundCard records and plays at the operating system’s default configuration. Particularly on laptops, this configuration might have extreme latencies, up to multiple seconds.

In order to request lower latencies, pass a blocksize to player or recorder. This tells the operating system your desired latency, and it will try to honor your request as best it can. On Windows/WASAPI, setting exclusive_mode=True might help, too (this is currently experimental).

Another source of latency is in the record function, which buffers output up to the requested numframes. In general, for optimal latency, you should use a numframes significantly lower than the blocksize above, maybe by a factor of two or four.

To get the audio data as quickly as absolutely possible, you can use numframes=None, which will return whatever audio data is available right now, without any buffering. Note that this might receive different numbers of frames each time.

With the above settings, block sizes of 256 samples or ten milliseconds are usually no problem. The total latency of playback and recording is dependent on how these buffers are handled by the operating system, though, and might be significantly higher.

Additionally, it might help to experiment with advice from here: https://askubuntu.com/questions/707171/how-can-i-fix-choppy-audio and edit your /etc/pulse/default.pa file to replace the line saying

load-module module-udev-detect

with

load-module module-udev-detect tsched=0

and then do not forget to restart pulseaudio with

pulseaudio -k

Channel Maps

Some professional sound cards have large numbers of channels. If you want to record or play only a subset of those channels, you can specify a channel map. For playback, a channel map of [0, 3, 4] will play three-channel audio data on the physical channels one, four, and five. For recording, a channel map of [0, 3, 4] will return three-channel audio data recorded from the physical channels one, four, and five.

In addition, pulseaudio/Linux defines channel -1 as the mono mix of all channels for both playback and recording. CoreAudio/macOS defines channel -1 as silence for both playback and recording.

Known Issues:

• Windows/WASAPI currently records garbage if you record only a single channel. The reason for this is yet unknown. Multi-channel and channel maps work, though.
• Windows/WASAPI silently ignores the blocksize in some cases. Apparently, it only supports variable block sizes in exclusive mode.
• Error messages often report some internal CFFI/backend errors. This will be improved in the future.

Changelog

• 2018-04-25 implements fixed block sizes when recording (thank you, Pariente Manuel!)
• 2018-05-10 adds a test suite and various fixes for Windows
• 2018-05-11 various fixes for macOS
• 2018-06-27 Adds latency property to Linux/pulseaudio (Thank you, Pariente Manuel!)
• 2018-07-17 adds loopback support for Windows (Thank you, Jan Leskovec!)
• 2018-10-16 adds bug fix for IPython on Windows (Thank you, Sebastian Michel!)
• 2018-11-28 adds Sphinx/Readthedocs documentation
• 2019-03-25 adds support for Python 3.5 (Thank you, Daniel R. Kumor!)
• 2019-04-29 adds experimental support for exclusive mode on Windows
• 2019-05-13 fixes sample rate conversion on macOS
• 2019-05-15 fixes silence recording on macOS
• 2019-06-11 fixes exception when monitoring default device on Linux (Thank you, Inti Pelupessy!)
• 2019-06-18 fixes crash when opening many streams on Linux
• 2019-08-23 fixes attribute error when accessing stream state on Linux (Thank you, Davíð Sindri Pétursson!)
• 2019-10-08 fixes inconsistent dtypes when recording on Linux
• 2020-01-06 fixes silent recordings on Windows
• 2020-04-28 get and set the pulseaudio program name on Linux (Thank you, Philipp A.!)
• 2020-05-14 fixes error with unicode soundcard names on Windows (Thank you, BAKEZQ!)
• 2020-05-18 adds support for pyinstaller (v4) (Thank you, Bob Thomas!)
• 2020-05-19 adds compatibility with Windows 7 (Thank you, demberto!)

API Documentation

The following functions are your entry point to the SoundCard library. All of them return instances of soundcard._Speaker or soundcard._Microphone. These are lightweight objects that reference a sound card, but don’t do any actual work.

Since audio hardware can change frequently when new audio devices are plugged in or out, it is a good idea to always retrieve new instances of soundcard._Speaker and soundcard._Microphone instead of keeping old references around for a long time.

soundcard.all_microphones(include_loopback=False, exclude_monitors=True)

A list of all connected microphones.

By default, this does not include loopbacks (virtual microphones that record the output of a speaker).

Parameters:

• include_loopback (bool) – allow recording of speaker outputs
• exclude_monitors (bool) – deprecated version of include_loopback

Returns:

microphones

Return type:

list(_Microphone)

soundcard.all_speakers()

A list of all connected speakers.

Returns:speakers Return type:list(_Speaker)

soundcard.cffi

soundcard.default_microphone()

The default microphone of the system.

Returns:microphone Return type:_Microphone

soundcard.default_speaker()

The default speaker of the system.

Returns:speaker Return type:_Speaker

soundcard.get_microphone(id, include_loopback=False, exclude_monitors=True)

Get a specific microphone by a variety of means.

By default, this does not include loopbacks (virtual microphones that record the output of a speaker).

Parameters:

• id (int or str) – can be an int index, a backend id, a substring of the speaker name, or a fuzzy-matched pattern for the speaker name.
• include_loopback (bool) – allow recording of speaker outputs
• exclude_monitors (bool) – deprecated version of include_loopback

Returns:

microphone

Return type:

_Microphone

soundcard.get_name()

Get application name.

Note

Currently only works on Linux.

Returns:name Return type:str

soundcard.get_speaker(id)

Get a specific speaker by a variety of means.

Parameters:id (int or str) – can be an int index, a backend id, a substring of the speaker name, or a fuzzy-matched pattern for the speaker name. Returns:speaker Return type:_Speaker

soundcard.set_name(name)

Set application name.

Note

Currently only works on Linux.

Parameters:name (str) – The application using the soundcard will be identified by the OS using this name.

Sound Card Handles

The following classes are lightweight references to sound cards. They are not meant to be constructed by hand, but are returned by default_speaker(), get_speaker(), all_speakers(), default_microphone(), get_microphone(), all_microphones().

The real work of interfacing with the hardware and playing or recording audio is delegated to the _Player and _Recorder objects, which are created by _Speaker.play(), _Speaker.player(), _Microphone.record(), and _Microphone.recorder().

class soundcard._Speaker(*, id)

A soundcard output. Can be used to play audio.

Use the play() method to play one piece of audio, or use the player() method to get a context manager for playing continuous audio.

Multiple calls to play() play immediately and concurrently, while the player() schedules multiple pieces of audio one after another.

channels

Either the number of channels, or a list of channel indices. Index -1 is the mono mixture of all channels, and subsequent numbers are channel numbers (left, right, center, …)

Type:int or list(int)

id

A backend-dependent unique ID.

Type:object

name

The human-readable name of the soundcard.

Type:str

play(data, samplerate, channels=None, blocksize=None)

Play some audio data.

Parameters:

• data (numpy array) – The audio data to play. Must be a frames x channels Numpy array.
• samplerate (int) – The desired sampling rate in Hz
• channels ({int, list(int)}, optional) – Play on these channels. For example, [0, 3] will play stereo data on the physical channels one and four. Defaults to use all available channels. On Linux, channel -1 is the mono mix of all channels. On macOS, channel -1 is silence.
• blocksize (int) – Will play this many samples at a time. Choose a lower block size for lower latency and more CPU usage.

player(samplerate, channels=None, blocksize=None)

Create Player for playing audio.

Parameters:

• samplerate (int) – The desired sampling rate in Hz
• channels ({int, list(int)}, optional) – Play on these channels. For example, [0, 3] will play stereo data on the physical channels one and four. Defaults to use all available channels. On Linux, channel -1 is the mono mix of all channels. On macOS, channel -1 is silence.
• blocksize (int) – Will play this many samples at a time. Choose a lower block size for lower latency and more CPU usage.
• exclusive_mode (bool, optional) – Windows only: open sound card in exclusive mode, which might be necessary for short block lengths or high sample rates or optimal performance. Default is False.

Returns:

player

Return type:

_Player

class soundcard._Microphone(*, id)

A soundcard input. Can be used to record audio.

Use the record() method to record one piece of audio, or use the recorder() method to get a context manager for recording continuous audio.

Multiple calls to record() record immediately and concurrently, while the recorder() schedules multiple pieces of audio to be recorded one after another.

channels

Either the number of channels, or a list of channel indices. Index -1 is the mono mixture of all channels, and subsequent numbers are channel numbers (left, right, center, …)

Type:int or list(int)

id

A backend-dependent unique ID.

Type:object

isloopback

Whether this microphone is recording a speaker.

Type:bool

name

The human-readable name of the soundcard.

Type:str

record(numframes, samplerate, channels=None, blocksize=None)

Record some audio data.

Parameters:

• numframes (int) – The number of frames to record.
• samplerate (int) – The desired sampling rate in Hz
• channels ({int, list(int)}, optional) – Record on these channels. For example, [0, 3] will record stereo data from the physical channels one and four. Defaults to use all available channels. On Linux, channel -1 is the mono mix of all channels. On macOS, channel -1 is silence.
• blocksize (int) – Will record this many samples at a time. Choose a lower block size for lower latency and more CPU usage.

Returns:

data – The recorded audio data. Will be a frames x channels Numpy array.

Return type:

numpy array

recorder(samplerate, channels=None, blocksize=None)

Create Recorder for recording audio.

Parameters:

• samplerate (int) – The desired sampling rate in Hz
• channels ({int, list(int)}, optional) – Record on these channels. For example, [0, 3] will record stereo data from the physical channels one and four. Defaults to use all available channels. On Linux, channel -1 is the mono mix of all channels. On macOS, channel -1 is silence.
• blocksize (int) – Will record this many samples at a time. Choose a lower block size for lower latency and more CPU usage.
• exclusive_mode (bool, optional) – Windows only: open sound card in exclusive mode, which might be necessary for short block lengths or high sample rates or optimal performance. Default is False.

Returns:

recorder

Return type:

_Recorder

Sound Streams

The following classes are the context managers that do the heavy lifting of interacting with the backend drivers and sound cards. They are not meant to be constructed by hand, but are returned by _Speaker.player(), and _Microphone.recorder().

class soundcard._Player(id, samplerate, channels, blocksize=None, name='outputstream')

A context manager for an active output stream.

Audio playback is available as soon as the context manager is entered. Audio data can be played using the play() method. Successive calls to play() will queue up the audio one piece after another. If no audio is queued up, this will play silence.

This context manager can only be entered once, and can not be used after it is closed.

latency

Latency of the stream in seconds (only available on Linux)

Type:float

play(data)

Play some audio data.

Internally, all data is handled as float32 and with the appropriate number of channels. For maximum performance, provide data as a frames × channels float32 numpy array.

If single-channel or one-dimensional data is given, this data will be played on all available channels.

This function will return before all data has been played, so that additional data can be provided for gapless playback. The amount of buffering can be controlled through the blocksize of the player object.

If data is provided faster than it is played, later pieces will be queued up and played one after another.

Parameters:data (numpy array) – The audio data to play. Must be a frames x channels Numpy array.

class soundcard._Recorder(*args, **kwargs)

A context manager for an active input stream.

Audio recording is available as soon as the context manager is entered. Recorded audio data can be read using the record() method. If no audio data is available, record() will block until the requested amount of audio data has been recorded.

This context manager can only be entered once, and can not be used after it is closed.

flush()

Return the last pending chunk.

After using the record() method, this will return the last incomplete chunk and delete it.

Returns:data – The recorded audio data. Will be a frames x channels Numpy array. Return type:numpy array

latency

Latency of the stream in seconds (only available on Linux)

Type:float

record(numframes=None)

Record a block of audio data.

The data will be returned as a frames × channels float32 numpy array. This function will wait until numframes frames have been recorded. If numframes is given, it will return exactly numframes frames, and buffer the rest for later.

If numframes is None, it will return whatever the audio backend has available right now. Use this if latency must be kept to a minimum, but be aware that block sizes can change at the whims of the audio backend.

If using record() with numframes=None after using record() with a required numframes, the last buffered frame will be returned along with the new recorded block. (If you want to empty the last buffered frame instead, use flush())

Parameters:numframes (int, optional) – The number of frames to record. Returns:data – The recorded audio data. Will be a frames x channels Numpy array. Return type:numpy array

Indices and tables

• Index
• Module Index
• Search Page