Add framewise evaluation

evaluators/separation_eval.py

@@ -12,7 +12,6 @@
 import sys
 import os
 import glob
-import os
 import numpy as np
 import eval_utilities
 

mir_eval/separation.py

@@ -185,6 +185,67 @@ def bss_eval_sources(reference_sources, estimated_sources):
     return (sdr[idx], sir[idx], sar[idx], popt)
 
 
+def bss_eval_sources_framewise(
+    reference_sources, estimated_sources, win, hop
+):
+    """Framewise computation of bss_eval_sources
+
+    Examples
+    --------
+    >>> # reference_sources[n] should be an ndarray of samples of the
+    >>> # n'th reference source
+    >>> # estimated_sources[n] should be the same for the n'th estimated
+    >>> # source
+    >>> (sdr, sir, sar,
+    ...  perm) = mir_eval.separation.bss_eval_sources_framewise(
+             reference_sources,
+    ...      estimated_sources)
+
+    Parameters
+    ----------
+    reference_sources : np.ndarray, shape=(nsrc, nsampl)
+        matrix containing true sources
+    estimated_sources : np.ndarray, shape=(nsrc, nsampl)
+        matrix containing estimated sources
+    win : int
+        Window length
+    hop : int
+        Hop size
+
+    Returns
+    -------
+    sdr : np.ndarray, shape=(nsrc, nframes)
+        vector of Signal to Distortion Ratios (SDR)
+    sir : np.ndarray, shape=(nsrc, nframes)
+        vector of Source to Interference Ratios (SIR)
+    sar : np.ndarray, shape=(nsrc, nframes)
+        vector of Sources to Artifacts Ratios (SAR)
+    perm : np.ndarray, shape=(nsrc, nframes)
+        vector containing the best ordering of estimated sources in
+        the mean SIR sense (estimated source number perm[j] corresponds to
+        true source number j)
+
+    """
+    nsrc = reference_sources.shape[0]
+
+    nwin = int(
+        np.floor((reference_sources.shape[1] - win + hop) / hop)
+    )
+
+    SDR = np.empty((nsrc, nwin))
+    SIR = np.empty((nsrc, nwin))
+    SAR = np.empty((nsrc, nwin))
+    perm = np.empty((nsrc, nwin))
+
+    for k in range(nwin):
+        K = slice(k * hop, k * hop + win)
+        SDR[:, k], SIR[:, k], SAR[:, k], perm[:, k] = bss_eval_sources(
+            reference_sources[:, K], estimated_sources[:, K]
+        )
+
+    return SDR, SIR, SAR, perm
+
+
 def _bss_decomp_mtifilt(reference_sources, estimated_source, j, flen):
     """Decomposition of an estimated source image into four components
     representing respectively the true source image, spatial (or filtering)
@@ -359,9 +420,12 @@ def evaluate(reference_sources, estimated_sources, **kwargs):
     # Compute all the metrics
     scores = collections.OrderedDict()
 
-    sdr, sir, sar, perm = util.filter_kwargs(bss_eval_sources,
-                                             reference_sources,
-                                             estimated_sources, **kwargs)
+    sdr, sir, sar, perm = util.filter_kwargs(
+        bss_eval_sources,
+        reference_sources,
+        estimated_sources,
+        **kwargs
+    )
 
     scores['Source to Distortion'] = sdr.tolist()
     scores['Source to Interference'] = sir.tolist()