pypi / librosa
librosa
What changed in librosa 1, from its official release notes, upgrade guides and advisories. Sources checked 2026-10-24.
16 breaking4 deprecations24 new APIs and options2 performance7 behavior and defaults6 fixes and other
See which of these apply to your code, ranked, in your agent:
npx @unstale/cli initlibrosa 159 changes
Breaking changes 16
When ref is a function, the default operation is over trailing axes rather than the entire array, preserving channel independence; axes=None restores the 0.11 behavior.
The audioread backend for loading audio files has been removed.
Removes filters.constant_q in favor of wavelet.
Removes filters.constant_q_lengths in favor of wavelet_lengths.
Removes get_fftlib.
display.wavef0 was added; it was previously named melodyplot.
The default quantization was removed, changing the function's default behavior.
Removes the filename argument in favor of path.
Removes the res_type argument.
When ref is a function, the default operation is over trailing axes rather than the entire array, preserving channel independence; axes=None restores the 0.11 behavior.
Removes the win_length parameter.
The minimum supported Python version is now 3.12.
Functions that previously returned scipy.sparse.spmatrix objects now return scipy.sparse.sparse_array objects.
Removes set_fftlib.
Removes the x_axis parameter in favor of axis.
Removes the win_length parameter.
Deprecations 4
Deprecates librosa.display.cmap in favor of librosa.display.infer_cmap.
Deprecates acceptance of hop_length and n_fft parameters; they will no longer be accepted in a future release.
Deprecates random_state parameters as PRNG interfaces migrate to rng.
Deprecates seed parameters as PRNG interfaces migrate to rng.
New APIs and options 24
Adds a helper for constructing colorbars with appropriate labeling for decibel-valued data.
Adds a helper for constructing colorbars with appropriate labeling for angle-valued data.
Adds a helper for applying path effects such as outlines or shadows to matplotlib artists so they are more visible over spectrogram displays.
Added the cmap_cyclic option to specshow.
Added the vscale option to specshow.
An inverted waveshow display was added.
The Metrogram Transform was added.
A hybrid tempogram implementation was added.
loadx was added, with a follow-up change to provide a more helpful failure mode.
Stream resampling was added.
Added an out parameter for in-place downmixing.
Dynamic vectorization was added.
Summarizes relative energy at different meters over time and can be used to estimate a recording's time signature.
Adds a way to map an existing display routine over an array of matplotlib axes, sharing common parameters across subplots.
Additional oct3 display modes were added.
Standardizes PRNG seeding on the rng= interface across the package.
Signal mixing helpers were added.
Adds a flexible interface for mixing signals into different multidimensional array shapes.
Adds a flexible interface for mixing signals into different multidimensional array shapes.
Adds a flexible interface for mixing signals into different multidimensional array shapes.
Added conversion of a warping path to a step grid.
Wavebars were added.
Adds a waveform display that accepts a signal and fundamental-frequency sequence and produces a frequency-displaced plot.
Waveform displays add an inverted mode that applies color styling to the background rather than the signal.
Performance 2
Import times are faster after eager compilation of certain numba-optimized subroutines was removed.
By default, uses a sparsified transition matrix to skip low-likelihood transitions, substantially speeding up viterbi and related algorithms.
Behavior and default changes 7
The release is intended to stabilize the API from the long-running 0.x series.
n_bins is inferred automatically in cqt and vqt.
Multichannel dB support was added.
Peak picking by dynamic programming was added.
Phase vocoding gained variable-rate and non-iterative capabilities.
Adds on-the-fly sample-rate conversion like load; conversion is not enabled by default in this release.
Sample-rate conversion remains disabled by default for backward compatibility; the notes say the default will change in 1.1 and recommend making sr explicit.
Fixes and other changes 6
Boolean data is supported by BoundedNorm for specshow.
Audio input is now validated in feature.rms.
Loading audio with a negative offset was addressed.
Underflow during perceptual weighting now produces a warning.
Python 3.14 support was added.
RNG parameters were made SPEC07 compliant.