librosax

amplitude_to_db(S: Array, *, ref: float | Callable = 1.0, amin: float = 1e-05, top_db: float | None = 80.0) → floating[Any] | Array

Convert an amplitude spectrogram to decibel (dB) units.

This is equivalent to power_to_db(S**2, ref=ref**2, amin=amin**2, top_db=top_db), but is provided for convenience.

Parameters:

  • S – Input amplitude spectrogram.

  • ref – Reference value for scaling. If scalar, the amplitude |S| is scaled relative to ref: 20 * log10(S / ref). If callable, the reference value is computed as ref(S). Default is 1.0.

  • amin – Minimum threshold for input values. Default is 1e-5.

  • top_db – Threshold the output at top_db below the peak. Default is 80.0.

Returns:

dB-scaled spectrogram.

Return type:

jnp.ndarray

See also

power_to_db, db_to_amplitude

istft(stft_matrix: Array, hop_length: int = None, win_length: int = None, n_fft: int = None, window: str = 'hann', center: bool = True, length: int = None)

Compute the Inverse Short-Time Fourier Transform (ISTFT).

This function reconstructs a waveform from an STFT matrix using JAX’s scipy.signal.istft implementation.

Parameters:

  • stft_matrix – The STFT matrix from which to compute the inverse.

  • hop_length – Number of samples between successive frames. Default is win_length // 4.

  • win_length – Window size. Default is n_fft.

  • n_fft – FFT size. Default is (stft_matrix.shape[-2] - 1) * 2.

  • window – Window function type. Default is "hann".

  • center – If True, assumes the waveform was padded so that frames were centered. Default is True.

  • length – Target length for the reconstructed signal. If None, the entire signal is returned.

Returns:

Reconstructed time-domain signal.

Return type:

jnp.ndarray

Raises:

AssertionError – If center is False because the function is only tested for center=True.

power_to_db(x: Array, amin: float = 1e-10, top_db: float | None = 80.0, ref: float = 1.0) → Array

Convert a power spectrogram to decibel (dB) units.

This function is a JAX implementation of librosa.power_to_db.

Parameters:

  • x – Input power spectrogram.

  • amin – Minimum threshold for input values. Default is 1e-10.

  • top_db – Threshold the output at top_db below the peak. Default is 80.0.

  • ref – Reference value for scaling. Default is 1.0.

Returns:

dB-scaled spectrogram.

Return type:

jnp.ndarray

Raises:

librosa.util.exceptions.ParameterError – If top_db is negative.

stft(waveform: Array, n_fft: int, hop_length: int = None, win_length: int = None, window: str = 'hann', center: bool = True, pad_mode: str = 'constant')

Compute the Short-Time Fourier Transform (STFT) of a waveform.

This function computes the STFT of the given waveform using JAX’s scipy.signal.stft implementation.

Parameters:

  • waveform – Input signal waveform.

  • n_fft – FFT size.

  • hop_length – Number of samples between successive frames. Default is win_length // 4.

  • win_length – Window size. Default is n_fft.

  • window – Window function type. Default is "hann".

  • center – If True, the waveform is padded so that frames are centered. Default is True.

  • pad_mode – Padding mode for the waveform. Must be one of ["constant", "reflect"]. Default is "constant".

Returns:

Complex STFT matrix.

Return type:

jnp.ndarray

Raises:

AssertionError – If pad_mode is not one of ["constant", "reflect"].