from functools import lru_cache
from typing import List
import librosa
import numpy as np
from mugen import paths
from mugen.events import EventList, Event
class AudioEvent(Event):
"""
An event in some audio
"""
pass
class End(AudioEvent):
"""
The end of some audio
"""
pass
class Beat(AudioEvent):
"""
A beat in some audio
"""
pass
class WeakBeat(Beat):
"""
A weak beat in some audio
"""
pass
class Onset(AudioEvent):
"""
An onset in some audio
"""
pass
class Audio:
"""
Wraps the audio ouput from librosa, providing access to extra features
Attributes
----------
file
Loaded audio file
samples
Audio samples
sample_rate
Audio sample rate
duration
Audio duration (seconds)
"""
file: str
sample_rate: int
samples: np.ndarray
duration: float
def __init__(self, file: str, *, sample_rate: int = 44100):
"""
Parameters
----------
file
Audio file to load
"""
self.file = file
self.samples, self.sample_rate = librosa.load(file, sr=sample_rate)
self.duration = librosa.get_duration(y=self.samples, sr=self.sample_rate)
def __repr__(self):
filename = paths.filename_from_path(self.file)
return f'<Audio, file: {filename}, duration: {self.duration}>'
def beats(self, trim: bool = False) -> EventList:
"""
Gets beat events
Parameters
----------
trim
Label weak leading and trailing beats separately
Returns
-------
Detected beat events from the audio
"""
untrimmed_beats = self._beats()
untrimmed_beats = EventList([Beat(beat) for beat in untrimmed_beats], end=self.duration)
if not trim:
beats = untrimmed_beats
else:
trimmed_beats = self._beats(trim=True)
trimmed_leading_beats = [beat for beat in untrimmed_beats.locations if beat < trimmed_beats[0]]
trimmed_trailing_beats = [beat for beat in untrimmed_beats.locations if beat > trimmed_beats[-1]]
# Mark leading & trailing trimmed beats as weak beats
trimmed_beats = EventList([Beat(beat) for beat in trimmed_beats], end=self.duration)
trimmed_leading_beats = EventList([WeakBeat(beat) for beat in trimmed_leading_beats], end=self.duration)
trimmed_trailing_beats = EventList([WeakBeat(beat) for beat in trimmed_trailing_beats], end=self.duration)
beats = trimmed_leading_beats + trimmed_beats + trimmed_trailing_beats
return beats
@lru_cache(maxsize=None)
def _beats(self, trim: bool = False) -> List[float]:
"""
Gets beat locations using librosa's beat tracker
Parameters
----------
trim
Whether to discard weak beats
Returns
-------
Beat locations
"""
if trim:
tempo, beats = librosa.beat.beat_track(y=self.samples, sr=self.sample_rate, units='time',
trim=True)
else:
tempo, beats = librosa.beat.beat_track(y=self.samples, sr=self.sample_rate, units='time',
trim=False)
return beats
def onsets(self, backtrack: bool = False) -> EventList:
"""
Gets onset events
Parameters
----------
backtrack
Shift onset events back to the nearest local minimum of energy
Returns
-------
Detected onset events from the audio
"""
if not backtrack:
onsets = self._onsets()
else:
onsets = self._onsets(backtrack=True)
onsets = EventList([Onset(onset) for onset in onsets], end=self.duration)
return onsets
@lru_cache(maxsize=None)
def _onsets(self, backtrack: bool = False):
"""
Gets onset locations using librosa's onset detector.
Parameters
----------
backtrack
Whether to shift onset events back to the nearest local minimum of energy
Returns
-------
Onset locations
"""
if backtrack:
onsets = librosa.beat.onset.onset_detect(y=self.samples, sr=self.sample_rate, units='time', backtrack=True)
else:
onsets = librosa.beat.onset.onset_detect(y=self.samples, sr=self.sample_rate, units='time', backtrack=False)
return onsets
