v1.3.1

A bugfix release: CPU inference with onnx, pt and pb is several times faster at the defaults, long MP3 files no longer slow the run down, and a handful of smaller defects are fixed.

Bugfixes

• CPU inference with onnx, pt and pb is several times faster at the defaults. Every worker used all cores at once; a default 3.0 onnx session is now about 8 times faster on an 8-core machine. by @putdanil in #120
• Long MP3 files no longer slow down with their length. A 60 min MP3 is about 7 times faster. by @putdanil in #119
• Loading a second result from .npz no longer replaces every result loaded earlier in the process. by @Josef-Haupt in #116
• GeoPredictionResult.latitude and .longitude are floats, not coordinates truncated to integers. by @Josef-Haupt in #116
• The geo models respect half_precision=True instead of always returning float32 scores. by @Josef-Haupt in #116
• encode(..) and encode_arrays(..) with show_stats no longer raise a TypeError after inference. by @Josef-Haupt in #116
• An input with more segments than the pipeline can address cancels the run instead of passing as valid with zero segments. by @Josef-Haupt in #116
• An audio array with more than two dimensions raises a ValueError instead of a bare AssertionError. by @Josef-Haupt in #117
• show_stats="benchmark" reports speed_worker_xrt_max, which was always 0.0. by @Josef-Haupt in #117

Full Changelog: v1.3.0...v1.3.1