iddlmZddlmZmZmZmZddlmZm Z ddl m Z ddl m Z ddlmZddlmZddlmZmZd d giZesd gZGd d eZy ))Sequence)AnyCallableOptionalUnion)Tensortensor)Literal)permutation_invariant_training)Metric)_MATPLOTLIB_AVAILABLE)_AX_TYPE_PLOT_OUT_TYPEPermutationInvariantTrainingpitz!PermutationInvariantTraining.plotc eZdZUdZdZeed<dZeed<eed<eed<dZ e e ed <dZ e e ed < dd e d ed deddeddf fd ZdededdfdZdefdZddeeeedfde edefdZxZS)raQCalculate `Permutation invariant training`_ (PIT). This metric can evaluate models for speaker independent multi-talker speech separation in a permutation invariant way. As input to ``forward`` and ``update`` the metric accepts the following input - ``preds`` (:class:`~torch.Tensor`): float tensor with shape ``(batch_size,num_speakers,...)`` - ``target`` (:class:`~torch.Tensor`): float tensor with shape ``(batch_size,num_speakers,...)`` As output of `forward` and `compute` the metric returns the following output - ``pesq`` (:class:`~torch.Tensor`): float scalar tensor with average PESQ value over samples Args: metric_func: a metric function accept a batch of target and estimate. if `mode`==`'speaker-wise'`, then ``metric_func(preds[:, i, ...], target[:, j, ...])`` is called and expected to return a batch of metric tensors ``(batch,)``; if `mode`==`'permutation-wise'`, then ``metric_func(preds[:, p, ...], target[:, :, ...])`` is called, where `p` is one possible permutation, e.g. [0,1] or [1,0] for 2-speaker case, and expected to return a batch of metric tensors ``(batch,)``; mode: can be `'speaker-wise'` or `'permutation-wise'`. eval_func: the function to find the best permutation, can be 'min' or 'max', i.e. the smaller the better or the larger the better. kwargs: Additional keyword arguments for either the ``metric_func`` or distributed communication, see :ref:`Metric kwargs` for more info. Example: >>> from torch import randn >>> from torchmetrics.audio import PermutationInvariantTraining >>> from torchmetrics.functional.audio import scale_invariant_signal_noise_ratio >>> preds = randn(3, 2, 5) # [batch, spk, time] >>> target = randn(3, 2, 5) # [batch, spk, time] >>> pit = PermutationInvariantTraining(scale_invariant_signal_noise_ratio, ... mode="speaker-wise", eval_func="max") >>> pit(preds, target) tensor(-2.1065) Ffull_state_updateTis_differentiablesum_pit_metrictotalNplot_lower_boundplot_upper_bound metric_funcmode) speaker-wisezpermutation-wise eval_func)maxminkwargsreturnc <|jdd|jdd|jddd}t|d i|||_||_||_||_|jdtdd |jd td d y) Ndist_sync_on_stepF process_group dist_sync_fn)r"r#r$rgsum)defaultdist_reduce_fxrr) popsuper__init__rrrr add_stater )selfrrrr base_kwargs __class__s k/Volumes/fast/ai/experiments/voice-extract-mac/.venv/lib/python3.12/site-packages/torchmetrics/audio/pit.pyr+z%PermutationInvariantTraining.__init__Ts"(,?!G#ZZ>"JJ~t<' ';'& "  'US wq %Hpredstargetct|||j|j|jfi|jd}|xj |j z c_|xj|jz c_y)z*Update state with predictions and targets.rN) r rrrrrr%rnumel)r-r2r3 pit_metrics r0updatez#PermutationInvariantTraining.updateisj3 64++TYY JN++   z~~// j&&(( r1c4|j|jz S)zCompute metric.)rr)r-s r0computez$PermutationInvariantTraining.computers""TZZ//r1valaxc&|j||S)aPlot a single or multiple values from the metric. Args: val: Either a single result from calling `metric.forward` or `metric.compute` or a list of these results. If no value is provided, will automatically call `metric.compute` and plot that result. ax: An matplotlib axis object. If provided will add plot to that axis Returns: Figure and Axes object Raises: ModuleNotFoundError: If `matplotlib` is not installed .. plot:: :scale: 75 >>> # Example plotting a single value >>> import torch >>> from torchmetrics.audio import PermutationInvariantTraining >>> from torchmetrics.functional.audio import scale_invariant_signal_noise_ratio >>> preds = torch.randn(3, 2, 5) # [batch, spk, time] >>> target = torch.randn(3, 2, 5) # [batch, spk, time] >>> metric = PermutationInvariantTraining(scale_invariant_signal_noise_ratio, ... mode="speaker-wise", eval_func="max") >>> metric.update(preds, target) >>> fig_, ax_ = metric.plot() .. plot:: :scale: 75 >>> # Example plotting multiple values >>> import torch >>> from torchmetrics.audio import PermutationInvariantTraining >>> from torchmetrics.functional.audio import scale_invariant_signal_noise_ratio >>> preds = torch.randn(3, 2, 5) # [batch, spk, time] >>> target = torch.randn(3, 2, 5) # [batch, spk, time] >>> metric = PermutationInvariantTraining(scale_invariant_signal_noise_ratio, ... mode="speaker-wise", eval_func="max") >>> values = [ ] >>> for _ in range(10): ... values.append(metric(preds, target)) >>> fig_, ax_ = metric.plot(values) )_plot)r-r:r;s r0plotz!PermutationInvariantTraining.plotvs\zz#r""r1)rr)NN)__name__ __module__ __qualname____doc__rbool__annotations__rrrrfloatrrr rr+r7r9rrrrr> __classcell__)r/s@r0rrs+Z$t#"t" M(,huo,(,huo, =K+0 II89I<( I  I  I*)F)F)t)00.#fhv&6<=.#(S[J\.#hv.#r1N)collections.abcrtypingrrrrtorchrr typing_extensionsr !torchmetrics.functional.audio.pitr torchmetrics.metricr torchmetrics.utilities.importsr torchmetrics.utilities.plotrr__doctest_requires____doctest_skip__rr(r1r0rQsI%11 %L&@@6@;<E#6E#r1