diart.mapping#

Module Contents#

Classes#

MappingMatrixObjective

Helper class that provides a standard way to create an ABC using

MinimizationObjective

Helper class that provides a standard way to create an ABC using

MaximizationObjective

Helper class that provides a standard way to create an ABC using

SpeakerMapBuilder

SpeakerMap

class diart.mapping.MappingMatrixObjective#

Bases: abc.ABC

Helper class that provides a standard way to create an ABC using inheritance.

property invalid_value: float#
Return type:

float

abstract property maximize: bool#
Return type:

bool

abstract property best_possible_value: float#
Return type:

float

abstract property best_value_fn: Callable#
Return type:

Callable

invalid_tensor(shape)#
Parameters:

shape (Union[Tuple, int]) –

Return type:

numpy.ndarray

optimal_assignments(matrix)#
Parameters:

matrix (numpy.ndarray) –

Return type:

List[int]

mapped_indices(matrix, axis)#
Parameters:

  • matrix (numpy.ndarray) –

  • axis (int) –

Return type:

List[int]

hard_speaker_map(num_src, num_tgt, assignments)#

Create a hard map object where the highest cost is put everywhere except on hard assignments from assignments.

Parameters:

  • num_src (int) – Number of source speakers

  • num_tgt (int) – Number of target speakers

  • assignments (Iterable[Tuple[int, int]]) – An iterable of tuples with two elements having the first element as the source speaker and the second element as the target speaker

Return type:

SpeakerMap

class diart.mapping.MinimizationObjective#

Bases: MappingMatrixObjective

Helper class that provides a standard way to create an ABC using inheritance.

property maximize: bool#
Return type:

bool

property best_possible_value: float#
Return type:

float

property best_value_fn: Callable#
Return type:

Callable

class diart.mapping.MaximizationObjective(max_value=1)#

Bases: MappingMatrixObjective

Helper class that provides a standard way to create an ABC using inheritance.

Parameters:

max_value (float) –

property maximize: bool#
Return type:

bool

property best_possible_value: float#
Return type:

float

property best_value_fn: Callable#
Return type:

Callable

class diart.mapping.SpeakerMapBuilder#
static hard_map(shape, assignments, maximize)#

Create a SpeakerMap object based on the given assignments. This is a “hard” map, meaning that the highest cost is put everywhere except on hard assignments from assignments.

Parameters:

  • shape (Tuple[int, int])) – Shape of the mapping matrix

  • assignments (Iterable[Tuple[int, int]]) – An iterable of tuples with two elements having the first element as the source speaker and the second element as the target speaker

  • maximize (bool) – whether to use scores where higher is better (true) or where lower is better (false)

Return type:

SpeakerMap

static correlation(scores1, scores2)#
Parameters:

  • scores1 (numpy.ndarray) –

  • scores2 (numpy.ndarray) –

Return type:

SpeakerMap

static mse(scores1, scores2)#
Parameters:

  • scores1 (numpy.ndarray) –

  • scores2 (numpy.ndarray) –

Return type:

SpeakerMap

static mae(scores1, scores2)#
Parameters:

  • scores1 (numpy.ndarray) –

  • scores2 (numpy.ndarray) –

Return type:

SpeakerMap

static dist(embeddings1, embeddings2, metric='cosine')#
Parameters:

  • embeddings1 (numpy.ndarray) –

  • embeddings2 (numpy.ndarray) –

  • metric (Text) –

Return type:

SpeakerMap

class diart.mapping.SpeakerMap(mapping_matrix, objective)#
Parameters:
property _raw_optimal_assignments: List[int]#
Return type:

List[int]

property shape: Tuple[int, int]#
Return type:

Tuple[int, int]

__len__()#
__add__(other)#
Parameters:

other (SpeakerMap) –

Return type:

SpeakerMap

_strict_check_valid(src, tgt)#
Parameters:

  • src (int) –

  • tgt (int) –

Return type:

bool

_loose_check_valid(src, tgt)#
Parameters:

  • src (int) –

  • tgt (int) –

Return type:

bool

valid_assignments(strict=False, as_array=False)#
Parameters:

  • strict (bool) –

  • as_array (bool) –

Return type:

Union[Tuple[List[int], List[int]], Tuple[numpy.ndarray, numpy.ndarray]]

to_dict(strict=False)#
Parameters:

strict (bool) –

Return type:

Dict[int, int]

to_inverse_dict(strict=False)#
Parameters:

strict (bool) –

Return type:

Dict[int, int]

is_source_speaker_mapped(source_speaker)#
Parameters:

source_speaker (int) –

Return type:

bool

is_target_speaker_mapped(target_speaker)#
Parameters:

target_speaker (int) –

Return type:

bool

set_source_speaker(src_speaker, tgt_speaker)#
Parameters:

tgt_speaker (int) –

unmap_source_speaker(src_speaker)#
Parameters:

src_speaker (int) –

unmap_threshold(threshold)#
Parameters:

threshold (float) –

Return type:

SpeakerMap

unmap_speakers(source_speakers=None, target_speakers=None)#
Parameters:

  • source_speakers (Optional[Union[List[int], numpy.ndarray]]) –

  • target_speakers (Optional[Union[List[int], numpy.ndarray]]) –

Return type:

SpeakerMap

compose(other)#

Let’s say that self is a mapping of source_speakers to intermediate_speakers and other is a mapping from intermediate_speakers to target_speakers.

Compose self with other to obtain a new mapping from source_speakers to target_speakers.

Parameters:

other (SpeakerMap) –

Return type:

SpeakerMap

union(other)#

self and other are two maps with the same dimensions. Return a new hard speaker map containing assignments in both maps.

An assignment from other is ignored if it is in conflict with a source or target speaker from self.

WARNING: The resulting map doesn’t preserve soft assignments because self and other might have different objectives.

Parameters:

other (SpeakerMap) – SpeakerMap Another speaker map

apply(source_scores)#

Apply this mapping to a score matrix of source speakers to obtain the same scores aligned to target speakers.

Parameters:

source_scores (SlidingWindowFeature, (num_frames, num_source_speakers)) – Source speaker scores per frame.

Returns:

projected_scores – Score matrix for target speakers.

Return type:

SlidingWindowFeature, (num_frames, num_target_speakers)