das.segment_utils#

Segment (syllable) utilities.

das.segment_utils.fill_gaps(labels: numpy.ndarray, gap_dur: int = 100) → numpy.ndarray[source]#

Fill short gaps in a sequence of labelled samples.

---111111-1111--- -> ---11111111111---

Parameters

  • labels (np.ndarray) – Sequence of labelled samples.

  • gap_dur (int, optional) – Minimal gap duration, in samples. Defaults to 100.

Returns

Labelled samples with short gaps filled.

Return type

np.ndarray

das.segment_utils.label_syllables_by_majority(labels: numpy.ndarray, onsets_seconds: numpy.ndarray, offsets_seconds: numpy.ndarray, samplerate: float) → Tuple[numpy.ndarray, numpy.ndarray][source]#

Label syllables by a majority vote.

Parameters

  • labels (np.ndarray) – Sequence of dirty, per sample, labels.

  • onsets_seconds (List[float]) – Onset of each syllable in labels, in seconds.

  • offsets_seconds (List[float]) – Offset of each syllable in labels, in seconds.

  • samplerate (float) – Samplerate of labels, in Hz.

Returns

Sequence of syllables, clean sequence of per-sample labels.

Return type

Tuple[np.ndarray, np.ndarray]

das.segment_utils.levenshtein(seq1: str, seq2: str) → float[source]#

Compute the Levenshtein edit distance between two strings.

Corresponds to the minimal number of insertions, deletions, and subsitutions required to transform seq1 into seq2.

Parameters

  • seq1 (str) –

  • seq2 (str) –

Returns

The Levenshtein distance between seq1 and seq1.

Return type

float

das.segment_utils.remove_short(labels: numpy.ndarray, min_len: int = 100) → numpy.ndarray[source]#

Remove short syllables from sequence of labelled samples.

---1111-1---1-- -> ---1111--------

Parameters

  • labels (np.ndarray) – Sequence of labelled samples.

  • min_len (int, optional) – Minimal segment (syllable) duration, in samples. Defaults to 100.

Returns

Labelled samples with short syllables removed.

Return type

np.ndarray

das.segment_utils.syllable_error_rate(true: str, pred: str) → float[source]#

Compute the Levenshtein edit distance normalized by length of true.

Parameters

  • true (str) – Ground truth labels for a sequence of syllables. For instance, ‘ABCDAAE’.

  • pred (str) – Predicted labels for a sequence of syllables.

Raises

TypeError – if either input is not a str

Returns

Levenshtein distance normalized by length of true.

Return type

float