#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
Onset detection
===============
.. autosummary::
:toctree: generated/
onset_detect
onset_backtrack
onset_strength
onset_strength_multi
"""
import numpy as np
import scipy
from ._cache import cache
from . import core
from . import util
from .util.exceptions import ParameterError
from .feature.spectral import melspectrogram
from typing import Any, Callable, Iterable, Optional, Union, Sequence
__all__ = ["onset_detect", "onset_strength", "onset_strength_multi", "onset_backtrack"]
[docs]def onset_detect(
*,
y: Optional[np.ndarray] = None,
sr: float = 22050,
onset_envelope: Optional[np.ndarray] = None,
hop_length: int = 512,
backtrack: bool = False,
energy: Optional[np.ndarray] = None,
units: str = "frames",
normalize: bool = True,
sparse: bool = True,
**kwargs: Any,
) -> np.ndarray:
"""Locate note onset events by picking peaks in an onset strength envelope.
The `peak_pick` parameters were chosen by large-scale hyper-parameter
optimization over the dataset provided by [#]_.
.. [#] https://github.com/CPJKU/onset_db
Parameters
----------
y : np.ndarray [shape=(..., n)]
audio time-series. Multi-channel is supported.
sr : number > 0 [scalar]
sampling rate of ``y``
onset_envelope : np.ndarray [shape=(..., m)]
(optional) pre-computed onset strength envelope
hop_length : int > 0 [scalar]
hop length (in samples)
units : {'frames', 'samples', 'time'}
The units to encode detected onset events in.
By default, 'frames' are used.
backtrack : bool
If ``True``, detected onset events are backtracked to the nearest
preceding minimum of ``energy``.
This is primarily useful when using onsets as slice points for segmentation.
.. note:: backtracking is only supported if ``sparse=True``.
energy : np.ndarray [shape=(m,)] (optional)
An energy function to use for backtracking detected onset events.
If none is provided, then ``onset_envelope`` is used.
normalize : bool
If ``True`` (default), normalize the onset envelope to have minimum of 0 and
maximum of 1 prior to detection. This is helpful for standardizing the
parameters of `librosa.util.peak_pick`.
Otherwise, the onset envelope is left unnormalized.
sparse : bool
If ``True`` (default), detections are returned as an array of frames,
samples, or time indices (as specified by ``units=``).
If ``False``, detections are encoded as a dense boolean array where
``onsets[n]`` is True if there's an onset at frame index ``n``.
.. note:: multi-channel input is only supported if ``sparse=False``.
**kwargs : additional keyword arguments
Additional parameters for peak picking.
See `librosa.util.peak_pick` for details.
Returns
-------
onsets : np.ndarray [shape=(n_onsets,) or onset_envelope.shape]
estimated positions of detected onsets, in whichever units
are specified. By default, frame indices.
If `sparse=False`, `onsets[..., n]` indicates an onset
detection at frame index `n`.
.. note::
If no onset strength could be detected, onset_detect returns
an empty array (sparse=True) or all-False array (sparse=False).
Raises
------
ParameterError
if neither ``y`` nor ``onsets`` are provided
or if ``units`` is not one of 'frames', 'samples', or 'time'
See Also
--------
onset_strength : compute onset strength per-frame
onset_backtrack : backtracking onset events
librosa.util.peak_pick : pick peaks from a time series
Examples
--------
Get onset times from a signal
>>> y, sr = librosa.load(librosa.ex('trumpet'))
>>> librosa.onset.onset_detect(y=y, sr=sr, units='time')
array([0.07 , 0.232, 0.395, 0.604, 0.743, 0.929, 1.045, 1.115,
1.416, 1.672, 1.881, 2.043, 2.206, 2.368, 2.554, 3.019])
Or use a pre-computed onset envelope
>>> o_env = librosa.onset.onset_strength(y=y, sr=sr)
>>> times = librosa.times_like(o_env, sr=sr)
>>> onset_frames = librosa.onset.onset_detect(onset_envelope=o_env, sr=sr)
>>> import matplotlib.pyplot as plt
>>> D = np.abs(librosa.stft(y))
>>> fig, ax = plt.subplots(nrows=2, sharex=True)
>>> librosa.display.specshow(librosa.amplitude_to_db(D, ref=np.max),
... x_axis='time', y_axis='log', ax=ax[0])
>>> ax[0].set(title='Power spectrogram')
>>> ax[0].label_outer()
>>> ax[1].plot(times, o_env, label='Onset strength')
>>> ax[1].vlines(times[onset_frames], 0, o_env.max(), color='r', alpha=0.9,
... linestyle='--', label='Onsets')
>>> ax[1].legend()
"""
# First, get the frame->beat strength profile if we don't already have one
if onset_envelope is None:
if y is None:
raise ParameterError("y or onset_envelope must be provided")
onset_envelope = onset_strength(y=y, sr=sr, hop_length=hop_length)
# Shift onset envelope up to be non-negative
# (a common normalization step to make the threshold more consistent)
if normalize:
# Normalize onset strength function to [0, 1] range
# Normalization is performed over the trailing axis
onset_envelope = onset_envelope - np.min(onset_envelope, keepdims=True, axis=-1)
# Mypy does not realize that oenv is not None by now
# Max-scale with safe division
onset_envelope /= np.max(onset_envelope, keepdims=True, axis=-1) + util.tiny(onset_envelope) # type: ignore
# help out mypy
assert onset_envelope is not None
# Do we have any onsets to grab?
if not onset_envelope.any() or not np.all(np.isfinite(onset_envelope)):
if sparse:
onsets = np.array([], dtype=int)
else:
onsets = np.zeros_like(onset_envelope, dtype=bool)
else:
# These parameter settings found by large-scale search
kwargs.setdefault("pre_max", 0.03 * sr // hop_length) # 30ms
kwargs.setdefault("post_max", 0.00 * sr // hop_length + 1) # 0ms
kwargs.setdefault("pre_avg", 0.10 * sr // hop_length) # 100ms
kwargs.setdefault("post_avg", 0.10 * sr // hop_length + 1) # 100ms
kwargs.setdefault("wait", 0.03 * sr // hop_length) # 30ms
kwargs.setdefault("delta", 0.07)
# Peak pick the onset envelope
onsets = util.peak_pick(onset_envelope, sparse=sparse, axis=-1, **kwargs)
# Optionally backtrack the events
if backtrack:
if not sparse:
raise ParameterError("onset backtracking is only supported if sparse=True")
if energy is None:
energy = onset_envelope
assert energy is not None
onsets = onset_backtrack(onsets, energy)
if sparse:
if units == "frames":
pass
elif units == "samples":
onsets = core.frames_to_samples(onsets, hop_length=hop_length)
elif units == "time":
onsets = core.frames_to_time(onsets, hop_length=hop_length, sr=sr)
else:
raise ParameterError(f"Invalid unit type: {units}")
return onsets
[docs]def onset_strength(
*,
y: Optional[np.ndarray] = None,
sr: float = 22050,
S: Optional[np.ndarray] = None,
lag: int = 1,
max_size: int = 1,
ref: Optional[np.ndarray] = None,
detrend: bool = False,
center: bool = True,
feature: Optional[Callable] = None,
aggregate: Optional[Union[Callable, bool]] = None,
**kwargs: Any,
) -> np.ndarray:
"""Compute a spectral flux onset strength envelope.
Onset strength at time ``t`` is determined by::
mean_f max(0, S[f, t] - ref[f, t - lag])
where ``ref`` is ``S`` after local max filtering along the frequency
axis [#]_.
By default, if a time series ``y`` is provided, S will be the
log-power Mel spectrogram.
.. [#] Böck, Sebastian, and Gerhard Widmer.
"Maximum filter vibrato suppression for onset detection."
16th International Conference on Digital Audio Effects,
Maynooth, Ireland. 2013.
Parameters
----------
y : np.ndarray [shape=(..., n)]
audio time-series. Multi-channel is supported.
sr : number > 0 [scalar]
sampling rate of ``y``
S : np.ndarray [shape=(..., d, m)]
pre-computed (log-power) spectrogram
lag : int > 0
time lag for computing differences
max_size : int > 0
size (in frequency bins) of the local max filter.
set to `1` to disable filtering.
ref : None or np.ndarray [shape=(..., d, m)]
An optional pre-computed reference spectrum, of the same shape as ``S``.
If not provided, it will be computed from ``S``.
If provided, it will override any local max filtering governed by ``max_size``.
detrend : bool [scalar]
Filter the onset strength to remove the DC component
center : bool [scalar]
Shift the onset function by ``n_fft // (2 * hop_length)`` frames.
This corresponds to using a centered frame analysis in the short-time Fourier
transform.
feature : function
Function for computing time-series features, eg, scaled spectrograms.
By default, uses `librosa.feature.melspectrogram` with ``fmax=sr/2``
aggregate : function
Aggregation function to use when combining onsets
at different frequency bins.
Default: `np.mean`
**kwargs : additional keyword arguments
Additional parameters to ``feature()``, if ``S`` is not provided.
Returns
-------
onset_envelope : np.ndarray [shape=(..., m,)]
vector containing the onset strength envelope.
If the input contains multiple channels, then onset envelope is computed for each channel.
Raises
------
ParameterError
if neither ``(y, sr)`` nor ``S`` are provided
or if ``lag`` or ``max_size`` are not positive integers
See Also
--------
onset_detect
onset_strength_multi
Examples
--------
First, load some audio and plot the spectrogram
>>> import matplotlib.pyplot as plt
>>> y, sr = librosa.load(librosa.ex('trumpet'), duration=3)
>>> D = np.abs(librosa.stft(y))
>>> times = librosa.times_like(D, sr=sr)
>>> fig, ax = plt.subplots(nrows=2, sharex=True)
>>> librosa.display.specshow(librosa.amplitude_to_db(D, ref=np.max),
... y_axis='log', x_axis='time', ax=ax[0])
>>> ax[0].set(title='Power spectrogram')
>>> ax[0].label_outer()
Construct a standard onset function
>>> onset_env = librosa.onset.onset_strength(y=y, sr=sr)
>>> ax[1].plot(times, 2 + onset_env / onset_env.max(), alpha=0.8,
... label='Mean (mel)')
Median aggregation, and custom mel options
>>> onset_env = librosa.onset.onset_strength(y=y, sr=sr,
... aggregate=np.median,
... fmax=8000, n_mels=256)
>>> ax[1].plot(times, 1 + onset_env / onset_env.max(), alpha=0.8,
... label='Median (custom mel)')
Constant-Q spectrogram instead of Mel
>>> C = np.abs(librosa.cqt(y=y, sr=sr))
>>> onset_env = librosa.onset.onset_strength(sr=sr, S=librosa.amplitude_to_db(C, ref=np.max))
>>> ax[1].plot(times, onset_env / onset_env.max(), alpha=0.8,
... label='Mean (CQT)')
>>> ax[1].legend()
>>> ax[1].set(ylabel='Normalized strength', yticks=[])
"""
if aggregate is False:
raise ParameterError(
f"aggregate parameter cannot be False when computing full-spectrum onset strength."
)
odf_all = onset_strength_multi(
y=y,
sr=sr,
S=S,
lag=lag,
max_size=max_size,
ref=ref,
detrend=detrend,
center=center,
feature=feature,
aggregate=aggregate,
channels=None,
**kwargs,
)
return odf_all[..., 0, :]
[docs]def onset_backtrack(events: np.ndarray, energy: np.ndarray) -> np.ndarray:
"""Backtrack detected onset events to the nearest preceding local
minimum of an energy function.
This function can be used to roll back the timing of detected onsets
from a detected peak amplitude to the preceding minimum.
This is most useful when using onsets to determine slice points for
segmentation, as described by [#]_.
.. [#] Jehan, Tristan.
"Creating music by listening"
Doctoral dissertation
Massachusetts Institute of Technology, 2005.
Parameters
----------
events : np.ndarray, dtype=int
List of onset event frame indices, as computed by `onset_detect`
energy : np.ndarray, shape=(m,)
An energy function
Returns
-------
events_backtracked : np.ndarray, shape=events.shape
The input events matched to nearest preceding minima of ``energy``.
Examples
--------
Backtrack the events using the onset envelope
>>> y, sr = librosa.load(librosa.ex('trumpet'), duration=3)
>>> oenv = librosa.onset.onset_strength(y=y, sr=sr)
>>> times = librosa.times_like(oenv, sr=sr)
>>> # Detect events without backtracking
>>> onset_raw = librosa.onset.onset_detect(onset_envelope=oenv,
... backtrack=False)
>>> onset_bt = librosa.onset.onset_backtrack(onset_raw, oenv)
Backtrack the events using the RMS values
>>> S = np.abs(librosa.stft(y=y))
>>> rms = librosa.feature.rms(S=S)
>>> onset_bt_rms = librosa.onset.onset_backtrack(onset_raw, rms[0])
Plot the results
>>> import matplotlib.pyplot as plt
>>> fig, ax = plt.subplots(nrows=3, sharex=True)
>>> librosa.display.specshow(librosa.amplitude_to_db(S, ref=np.max),
... y_axis='log', x_axis='time', ax=ax[0])
>>> ax[0].label_outer()
>>> ax[1].plot(times, oenv, label='Onset strength')
>>> ax[1].vlines(librosa.frames_to_time(onset_raw), 0, oenv.max(), label='Raw onsets')
>>> ax[1].vlines(librosa.frames_to_time(onset_bt), 0, oenv.max(), label='Backtracked', color='r')
>>> ax[1].legend()
>>> ax[1].label_outer()
>>> ax[2].plot(times, rms[0], label='RMS')
>>> ax[2].vlines(librosa.frames_to_time(onset_bt_rms), 0, rms.max(), label='Backtracked (RMS)', color='r')
>>> ax[2].legend()
"""
# Find points where energy is non-increasing
# all points: energy[i] <= energy[i-1]
# tail points: energy[i] < energy[i+1]
minima = np.flatnonzero((energy[1:-1] <= energy[:-2]) & (energy[1:-1] < energy[2:]))
# Pad on a 0, just in case we have onsets with no preceding minimum
# Shift by one to account for slicing in minima detection
minima = util.fix_frames(1 + minima, x_min=0)
# Only match going left from the detected events
results: np.ndarray = minima[util.match_events(events, minima, right=False)]
return results
[docs]@cache(level=30)
def onset_strength_multi(
*,
y: Optional[np.ndarray] = None,
sr: float = 22050,
S: Optional[np.ndarray] = None,
n_fft: int = 2048,
hop_length: int = 512,
lag: int = 1,
max_size: int = 1,
ref: Optional[np.ndarray] = None,
detrend: bool = False,
center: bool = True,
feature: Optional[Callable] = None,
aggregate: Optional[Union[Callable, bool]] = None,
channels: Optional[Union[Sequence[int], Sequence[slice]]] = None,
**kwargs: Any,
) -> np.ndarray:
"""Compute a spectral flux onset strength envelope across multiple channels.
Onset strength for channel ``i`` at time ``t`` is determined by::
mean_{f in channels[i]} max(0, S[f, t+1] - S[f, t])
Parameters
----------
y : np.ndarray [shape=(..., n,)]
audio time-series. Multi-channel is supported.
sr : number > 0 [scalar]
sampling rate of ``y``
S : np.ndarray [shape=(..., d, m)]
pre-computed (log-power) spectrogram
n_fft : int > 0 [scalar]
FFT window size for use in ``feature()`` if ``S`` is not provided.
hop_length : int > 0 [scalar]
hop length for use in ``feature()`` if ``S`` is not provided.
lag : int > 0
time lag for computing differences
max_size : int > 0
size (in frequency bins) of the local max filter.
set to `1` to disable filtering.
ref : None or np.ndarray [shape=(d, m)]
An optional pre-computed reference spectrum, of the same shape as ``S``.
If not provided, it will be computed from ``S``.
If provided, it will override any local max filtering governed by ``max_size``.
detrend : bool [scalar]
Filter the onset strength to remove the DC component
center : bool [scalar]
Shift the onset function by ``n_fft // (2 * hop_length)`` frames.
This corresponds to using a centered frame analysis in the short-time Fourier
transform.
feature : function
Function for computing time-series features, eg, scaled spectrograms.
By default, uses `librosa.feature.melspectrogram` with ``fmax=sr/2``
Must support arguments: ``y, sr, n_fft, hop_length``
aggregate : function or False
Aggregation function to use when combining onsets
at different frequency bins.
If ``False``, then no aggregation is performed.
Default: `np.mean`
channels : list or None
Array of channel boundaries or slice objects.
If `None`, then a single channel is generated to span all bands.
**kwargs : additional keyword arguments
Additional parameters to ``feature()``, if ``S`` is not provided.
Returns
-------
onset_envelope : np.ndarray [shape=(..., n_channels, m)]
array containing the onset strength envelope for each specified channel
Raises
------
ParameterError
if neither ``(y, sr)`` nor ``S`` are provided
See Also
--------
onset_strength
Notes
-----
This function caches at level 30.
Examples
--------
First, load some audio and plot the spectrogram
>>> import matplotlib.pyplot as plt
>>> y, sr = librosa.load(librosa.ex('choice'), duration=5)
>>> D = np.abs(librosa.stft(y))
>>> fig, ax = plt.subplots(nrows=2, sharex=True)
>>> img1 = librosa.display.specshow(librosa.amplitude_to_db(D, ref=np.max),
... y_axis='log', x_axis='time', ax=ax[0])
>>> ax[0].set(title='Power spectrogram')
>>> ax[0].label_outer()
>>> fig.colorbar(img1, ax=[ax[0]], format="%+2.f dB")
Construct a standard onset function over four sub-bands
>>> onset_subbands = librosa.onset.onset_strength_multi(y=y, sr=sr,
... channels=[0, 32, 64, 96, 128])
>>> img2 = librosa.display.specshow(onset_subbands, x_axis='time', ax=ax[1])
>>> ax[1].set(ylabel='Sub-bands', title='Sub-band onset strength')
>>> fig.colorbar(img2, ax=[ax[1]])
"""
if feature is None:
feature = melspectrogram
kwargs.setdefault("fmax", 0.5 * sr)
if aggregate is None:
aggregate = np.mean
if not util.is_positive_int(lag):
raise ParameterError(f"lag={lag} must be a positive integer")
if not util.is_positive_int(max_size):
raise ParameterError(f"max_size={max_size} must be a positive integer")
# First, compute mel spectrogram
if S is None:
S = np.abs(feature(y=y, sr=sr, n_fft=n_fft, hop_length=hop_length, **kwargs))
# Convert to dBs
S = core.power_to_db(S)
# Assertion to make type checking happy
assert S is not None
# Ensure that S is at least 2-d
S = np.atleast_2d(S)
# Compute the reference spectrogram.
# Efficiency hack: skip filtering step and pass by reference
# if max_size will produce a no-op.
if ref is None:
if max_size == 1:
ref = S
else:
ref = scipy.ndimage.maximum_filter1d(S, max_size, axis=-2)
elif ref.shape != S.shape:
raise ParameterError(
f"Reference spectrum shape {ref.shape} must match input spectrum {S.shape}"
)
# Compute difference to the reference, spaced by lag
onset_env = S[..., lag:] - ref[..., :-lag]
# Discard negatives (decreasing amplitude)
onset_env = np.maximum(0.0, onset_env)
# Aggregate within channels
pad = True
if channels is None:
channels = [slice(None)]
else:
pad = False
if callable(aggregate):
onset_env = util.sync(
onset_env, channels, aggregate=aggregate, pad=pad, axis=-2
)
# compensate for lag
pad_width = lag
if center:
# Counter-act framing effects. Shift the onsets by n_fft / hop_length
pad_width += n_fft // (2 * hop_length)
padding = [(0, 0) for _ in onset_env.shape]
padding[-1] = (int(pad_width), 0)
onset_env = np.pad(onset_env, padding, mode="constant")
# remove the DC component
if detrend:
onset_env = scipy.signal.lfilter([1.0, -1.0], [1.0, -0.99], onset_env, axis=-1)
# Trim to match the input duration
if center:
onset_env = onset_env[..., : S.shape[-1]]
return onset_env