Merge pull request #82 from readbeyond/devel

Fixing the index issue in dtw

Author: readbeyond

README.md

@@ -2,13 +2,13 @@
 
 **aeneas** is a Python/C library and a set of tools to automagically synchronize audio and text (aka forced alignment).
 
-* Version: 1.5.0
-* Date: 2016-04-02
+* Version: 1.5.0.2
+* Date: 2016-04-09
 * Developed by: [ReadBeyond](http://www.readbeyond.it/)
 * Lead Developer: [Alberto Pettarin](http://www.albertopettarin.it/)
 * License: the GNU Affero General Public License Version 3 (AGPL v3)
 * Contact: [[email protected]](mailto:[email protected])
-* Quick Links: [Home](http://www.readbeyond.it/aeneas/) - [GitHub](https://github.com/readbeyond/aeneas/) - [PyPI](https://pypi.python.org/pypi/aeneas/) - [Docs](http://www.readbeyond.it/aeneas/docs/) - [Tutorial](http://www.readbeyond.it/aeneas/docs/clitutorial.html) - [Mailing List](https://groups.google.com/d/forum/aeneas-forced-alignment) - [Web App](http://aeneasweb.org)
+* Quick Links: [Home](http://www.readbeyond.it/aeneas/) - [GitHub](https://github.com/readbeyond/aeneas/) - [PyPI](https://pypi.python.org/pypi/aeneas/) - [Docs](http://www.readbeyond.it/aeneas/docs/) - [Tutorial](http://www.readbeyond.it/aeneas/docs/clitutorial.html) - [Benchmark](https://readbeyond.github.io/aeneas-benchmark/) - [Mailing List](https://groups.google.com/d/forum/aeneas-forced-alignment) - [Web App](http://aeneasweb.org)
 
 
 ## Goal
@@ -206,10 +206,12 @@ which explains how to use the built-in command line tools.
   [https://groups.google.com/d/forum/aeneas-forced-alignment](https://groups.google.com/d/forum/aeneas-forced-alignment)
 * Changelog:
   [http://www.readbeyond.it/aeneas/docs/changelog.html](http://www.readbeyond.it/aeneas/docs/changelog.html)
-* High level description of how **aeneas** works:
+* High level description of how aeneas works:
   [HOWITWORKS](https://github.com/readbeyond/aeneas/blob/master/wiki/HOWITWORKS.md)
 * Development history:
   [HISTORY](https://github.com/readbeyond/aeneas/blob/master/wiki/HISTORY.md)
+* Benchmark suite:
+  [https://readbeyond.github.io/aeneas-benchmark/](https://readbeyond.github.io/aeneas-benchmark/)
 
 
 ## Supported Features

README.rst

@@ -4,8 +4,8 @@ aeneas
 **aeneas** is a Python/C library and a set of tools to automagically
 synchronize audio and text (aka forced alignment).
 
--  Version: 1.5.0
--  Date: 2016-04-02
+-  Version: 1.5.0.2
+-  Date: 2016-04-09
 -  Developed by: `ReadBeyond <http://www.readbeyond.it/>`__
 -  Lead Developer: `Alberto Pettarin <http://www.albertopettarin.it/>`__
 -  License: the GNU Affero General Public License Version 3 (AGPL v3)
@@ -15,7 +15,8 @@ synchronize audio and text (aka forced alignment).
    `PyPI <https://pypi.python.org/pypi/aeneas/>`__ -
    `Docs <http://www.readbeyond.it/aeneas/docs/>`__ -
    `Tutorial <http://www.readbeyond.it/aeneas/docs/clitutorial.html>`__
-   - `Mailing
+   - `Benchmark <https://readbeyond.github.io/aeneas-benchmark/>`__ -
+   `Mailing
    List <https://groups.google.com/d/forum/aeneas-forced-alignment>`__ -
    `Web App <http://aeneasweb.org>`__
 
@@ -208,10 +209,11 @@ Documentation and Support
 -  Mailing list:
    https://groups.google.com/d/forum/aeneas-forced-alignment
 -  Changelog: http://www.readbeyond.it/aeneas/docs/changelog.html
--  High level description of how **aeneas** works:
+-  High level description of how aeneas works:
    `HOWITWORKS <https://github.com/readbeyond/aeneas/blob/master/wiki/HOWITWORKS.md>`__
 -  Development history:
    `HISTORY <https://github.com/readbeyond/aeneas/blob/master/wiki/HISTORY.md>`__
+-  Benchmark suite: https://readbeyond.github.io/aeneas-benchmark/
 
 Supported Features
 ------------------

aeneas/dtw.py

@@ -222,11 +222,22 @@ def compute_boundaries(self, synt_anchors):
         # synt_anchors as in seconds, convert them in MFCC indices
         mws = self.rconf.mws
         anchor_indices = numpy.array([int(a[0] / mws) for a in synt_anchors])
+        #
         # right side sets the split point at the very beginning of "next" fragment
-        begin_indices = numpy.searchsorted(synt_indices, anchor_indices, side="right")
+        #
+        # NOTE clip() is needed since searchsorted() with side="right" might return
+        #      an index == len(synt_indices) == len(real_indices)
+        #      when the insertion point is past the last element of synt_indices
+        #      causing the fancy indexing real_indices[...] below might fail
+        begin_indices = numpy.clip(numpy.searchsorted(synt_indices, anchor_indices, side="right"), 0, len(synt_indices)-1)
         # first split must occur at zero
         begin_indices[0] = 0
+        #
         # map onto real indices, obtaining "default" boundary indices
+        #
+        # NOTE since len(synt_indices) == len(real_indices)
+        #      and because the numpy.clip() above, the fancy indexing is always valid
+        #
         boundary_indices = numpy.append(real_indices[begin_indices], self.real_wave_mfcc.tail_begin)
         self.log([u"Boundary indices: %d", len(boundary_indices)])
         self.log(u"Computing boundary indices... done")

docs/source/changelog.rst

@@ -1,6 +1,16 @@
 Changelog
 =========
 
+v1.5.0.2 (2016-04-09)
+---------------------
+
+#. Fix an issue in ``dtw`` with ``numpy.searchsorted`` returning an invalid index
+
+v1.5.0.1 (2016-04-03)
+---------------------
+
+#. Fix an issue with compiling C extensions on Windows
+
 v1.5.0 (2016-04-02)
 -------------------
 

setup.py

@@ -86,7 +86,7 @@
         "aeneas.extra": ["*.md"],
         "aeneas.tools": ["res/*", "*.md"]
     },
-    version="1.5.0.1",
+    version="1.5.0.2",
     description=SHORT_DESCRIPTION,
     author="Alberto Pettarin",
     author_email="[email protected]",