platypush/platypush/plugins/sound/core.py
2018-12-28 00:30:25 +01:00

380 lines
12 KiB
Python

"""
.. moduleauthor:: Fabio Manganiello <blacklight86@gmail.com>
"""
import enum
import json
import math
class WaveShape(enum.Enum):
SIN='sin'
SQUARE='square'
SAWTOOTH='sawtooth'
TRIANG='triang'
class Sound(object):
"""
Models a basic synthetic sound that can be played through an audio device
"""
STANDARD_A_FREQUENCY = 440.0
STANDARD_A_MIDI_NOTE = 69
_DEFAULT_BLOCKSIZE = 1024
_DEFAULT_SYNTH_BUFSIZE = 2
_DEFAULT_FILE_BUFSIZE = 20
_DEFAULT_SAMPLERATE = 44100
midi_note = None
frequency = None
phase = 0.0
gain = 1.0
duration = None
shape = None
def __init__(self, midi_note=midi_note, frequency=None, phase=phase,
gain=gain, duration=duration, shape=WaveShape.SIN,
A_frequency=STANDARD_A_FREQUENCY):
"""
You can construct a sound either from a MIDI note or a base frequency
:param midi_note: MIDI note code, see
https://newt.phys.unsw.edu.au/jw/graphics/notes.GIF
:type midi_note: int
:param frequency: Sound base frequency in Hz
:type frequency: float
:param phase: Wave phase shift as a multiple of pi (default: 0.0)
:type phase: float
:param gain: Note gain/volume between 0.0 and 1.0 (default: 1.0)
:type gain: float
:param duration: Note duration in seconds. Default: keep until
release/pause/stop
:type duration: float
:param shape: Wave shape. Possible values: "``sin``", "``square``",
"``sawtooth``" or "``triang``" (see :class:`WaveSound`).
Default: "``sin``"
:type shape: str
:param A_frequency: Reference A4 frequency (default: 440 Hz)
:type A_frequency: float
"""
if midi_note and frequency:
raise RuntimeError('Please specify either a MIDI note or a base ' +
'frequency')
if midi_note:
self.midi_note = midi_note
self.frequency = self.note_to_freq(midi_note=midi_note,
A_frequency=A_frequency)
elif frequency:
self.frequency = frequency
self.midi_note = self.freq_to_note(frequency=frequency,
A_frequency=A_frequency)
else:
raise RuntimeError('Please specify either a MIDI note or a base ' +
'frequency')
self.phase = phase
self.gain = gain
self.duration = duration
self.shape = WaveShape(shape)
@classmethod
def note_to_freq(cls, midi_note, A_frequency=STANDARD_A_FREQUENCY):
"""
Converts a MIDI note to its frequency in Hz
:param midi_note: MIDI note to convert
:type midi_note: int
:param A_frequency: Reference A4 frequency (default: 440 Hz)
:type A_frequency: float
"""
return (2.0 ** ((midi_note - cls.STANDARD_A_MIDI_NOTE) / 12.0)) \
* A_frequency
@classmethod
def freq_to_note(cls, frequency, A_frequency=STANDARD_A_FREQUENCY):
"""
Converts a frequency in Hz to its closest MIDI note
:param frequency: Frequency in Hz
:type midi_note: float
:param A_frequency: Reference A4 frequency (default: 440 Hz)
:type A_frequency: float
"""
# TODO return also the offset in % between the provided frequency
# and the standard MIDI note frequency
return int(12.0 * math.log(frequency/A_frequency, 2)
+ cls.STANDARD_A_MIDI_NOTE)
def get_wave(self, t_start=0., t_end=0., samplerate=_DEFAULT_SAMPLERATE):
"""
Get the wave binary data associated to this sound
:param t_start: Start offset for the wave in seconds. Default: 0
:type t_start: float
:param t_end: End offset for the wave in seconds. Default: 0
:type t_end: float
:param samplerate: Audio sample rate. Default: 44100 Hz
:type samplerate: int
:returns: A ``numpy.ndarray[(t_end-t_start)*samplerate, 1]``
with the raw float values
"""
import numpy as np
x = np.linspace(t_start, t_end, int((t_end-t_start)*samplerate))
x = x.reshape(len(x), 1)
if self.shape == WaveShape.SIN or self.shape == WaveShape.SQUARE:
wave = np.sin((2*np.pi*self.frequency*x) + np.pi*self.phase)
if self.shape == WaveShape.SQUARE:
wave[wave < 0] = -1
wave[wave >= 0] = 1
elif self.shape == WaveShape.SAWTOOTH or self.shape == WaveShape.TRIANG:
wave = 2 * (self.frequency*x -
np.floor(0.5 + self.frequency*x))
if self.shape == WaveShape.TRIANG:
wave = 2 * np.abs(wave) - 1
else:
raise RuntimeError('Unsupported wave shape: {}'.format(self.shape))
return self.gain * wave
def fft(self, t_start=0., t_end=0., samplerate=_DEFAULT_SAMPLERATE,
freq_range=None, freq_buckets=None):
"""
Get the real part of the Fourier transform associated to a time-bounded
sample of this sound
:param t_start: Start offset for the wave in seconds. Default: 0
:type t_start: float
:param t_end: End offset for the wave in seconds. Default: 0
:type t_end: float
:param samplerate: Audio sample rate. Default: 44100 Hz
:type samplerate: int
:param freq_range: FFT frequency range. Default: ``(0, samplerate/2)``
(see `Nyquist-Shannon sampling theorem <https://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem>`_)
:type freq_range: list or tuple with 2 int elements (range)
:param freq_buckets: Number of buckets to subdivide the frequency range.
Default: None
:type freq_buckets: int
:returns: A numpy.ndarray[freq_range,1] with the raw float values
"""
import numpy as np
if not freq_range:
freq_range = (0, int(samplerate/2))
wave = self.get_wave(t_start=t_start, t_end=t_end, samplerate=samplerate)
fft = np.fft.fft(wave.reshape(len(wave)))
fft = fft.real[freq_range[0]:freq_range[1]]
if freq_buckets is not None:
fft = np.histogram(fft, bins=freq_buckets)
return fft
def __iter__(self):
for attr in ['midi_note', 'frequency', 'gain', 'duration']:
yield (attr, getattr(self, attr))
def __str__(self):
return json.dumps(dict(self))
@classmethod
def build(cls, *args, **kwargs):
"""
Construct a sound object either from a JSON representation or a
key-value representation
"""
if args:
if isinstance(args[0], cls):
return args[0]
if isinstance(args[0], str):
kwargs = json.loads(args[0])
elif isinstance(args[0], dict):
kwargs = args[0]
if kwargs:
return Sound(**kwargs)
raise RuntimeError('Usage: {}'.format(__doc__))
class Mix(object):
"""
This class models a set of mixed :class:`Sound` instances that can be played
through an audio stream to an audio device
"""
_sounds = None
def __init__(self, *sounds):
self._sounds = []
for sound in sounds:
self.add(sound)
def __iter__(self):
for sound in self._sounds:
yield dict(sound)
def __str__(self):
return json.dumps(list(self))
def add(self, sound):
self._sounds.append(Sound.build(sound))
def remove(self, sound_index):
if sound_index >= len(self._sounds):
self.logger.error('No such sound index: {} in mix {}'.format(
sound_index, list(self)))
return
self._sounds.pop(sound_index)
def get_wave(self, t_start=0., t_end=0., normalize_range=(-1.0, 1.0),
on_clip='scale', samplerate=Sound._DEFAULT_SAMPLERATE):
"""
Get the wave binary data associated to this mix
:param t_start: Start offset for the wave in seconds. Default: 0
:type t_start: float
:param t_end: End offset for the wave in seconds. Default: 0
:type t_end: float
:param normalize_range: Normalization range. If set the gain values of the
wave will be normalized to fit into the specified range if it
"clips" above or below. Default: ``(-1.0, 1.0)``
:type normalize_range: list[float]
:param on_clip: Action to take on wave clipping if ``normalize_range``
is set. Possible values: "``scale``" (scale down the frame to remove
the clipping) or "``clip``" (saturate the values above/below range).
Default: "``scale``".
:type on_clip: str
:param samplerate: Audio sample rate. Default: 44100 Hz
:type samplerate: int
:returns: A numpy.ndarray[n,1] with the raw float values
"""
wave = None
for sound in self._sounds:
sound_wave = sound.get_wave(t_start=t_start, t_end=t_end,
samplerate=samplerate)
if wave is None:
wave = sound_wave
else:
wave += sound_wave
if normalize_range and len(wave):
scale_factor = (normalize_range[1]-normalize_range[0]) / \
(wave.max()-wave.min())
if scale_factor < 1.0: # Wave clipping
if on_clip == 'scale':
wave = scale_factor * wave
elif on_clip == 'clip':
wave[wave < normalize_range[0]] = normalize_range[0]
wave[wave > normalize_range[1]] = normalize_range[1]
else:
raise RuntimeError('Supported values for "on_clip": ' +
'"scale" or "clip"')
return wave
def fft(self, t_start=0., t_end=0., samplerate=Sound._DEFAULT_SAMPLERATE,
freq_range=None, freq_buckets=None):
"""
Get the real part of the Fourier transform associated to a time-bounded
sample of this mix
:param t_start: Start offset for the wave in seconds. Default: 0
:type t_start: float
:param t_end: End offset for the wave in seconds. Default: 0
:type t_end: float
:param samplerate: Audio sample rate. Default: 44100 Hz
:type samplerate: int
:param freq_range: FFT frequency range. Default: ``(0, samplerate/2)``
(see `Nyquist-Shannon sampling theorem <https://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem>`_)
:type freq_range: list or tuple with 2 int elements (range)
:param freq_buckets: Number of buckets to subdivide the frequency range.
Default: None
:type freq_buckets: int
:returns: A numpy.ndarray[freq_range,1] with the raw float values
"""
import numpy as np
if not freq_range:
freq_range = (0, int(samplerate/2))
wave = self.get_wave(t_start=t_start, t_end=t_end, samplerate=samplerate)
fft = np.fft.fft(wave.reshape(len(wave)))
fft = fft.real[freq_range[0]:freq_range[1]]
if freq_buckets is not None:
fft = np.histogram(fft, bins=freq_buckets)
return fft
def duration(self):
"""
:returns: The duration of the mix in seconds as duration of its longest
sample, or None if the mixed sample have no duration set
"""
duration = 0
for sound in self._sounds:
if sound.duration is None:
return None
duration = max(duration, sound.duration)
return duration
# vim:sw=4:ts=4:et: