diff --git a/pyannote/metrics/base.py b/pyannote/metrics/base.py
index 5d1dba7..49c22c1 100755
--- a/pyannote/metrics/base.py
+++ b/pyannote/metrics/base.py
@@ -25,6 +25,7 @@
 
 # AUTHORS
 # Hervé BREDIN - http://herve.niderb.fr
+import inspect
 from typing import List, Union, Optional, Set, Tuple
 
 import warnings
@@ -36,6 +37,49 @@
 from pyannote.metrics.types import Details, MetricComponents
 
 
+def clone(metric):
+    """Construct a new empty metric with the same parameters as `metric`.
+
+    Clone does a deep copy of the metric without actually copying any
+    results or accumulators. It returns a new metric (with the same
+    parameters) that has not yet been used to evaluate any data.
+
+    Parameters
+    ----------
+    metric : metric instance
+        The metric to be cloned.
+
+    Returns
+    -------
+    metric : object
+        The deep copy of the metric,
+    """
+    if hasattr(metric, '__pyannote_clone__') and not inspect.isclass(metric):
+        # If metric implements a custom cloning method, default to that.
+        return metric.__pyannote_clone__()
+    cls = metric.__class__
+    new_metric_params = metric.get_params()
+    new_metric = cls(**new_metric_params)
+
+    # Sanity check that all parameters were saved by the constructor without
+    # modification.
+    actual_params = new_metric.get_params()
+    for pname in new_metric_params:
+        error_msg = (
+            f'Cannot clone metric {repr(metric)}, as the constructor either '
+            f'does not set or modifies parameter {pname}.')
+        print(error_msg)
+        expected_param = new_metric_params[pname]
+        try:
+            actual_param = actual_params[pname]
+        except KeyError:
+            raise RuntimeError(error_msg())
+        if expected_param != actual_param:
+            raise RuntimeError(error_msg())
+
+    return new_metric
+
+
 class BaseMetric:
     """
     :class:`BaseMetric` is the base class for most pyannote evaluation metrics.
@@ -66,6 +110,33 @@ def __init__(self, **kwargs):
         self.components_: Set[str] = set(self.__class__.metric_components())
         self.reset()
 
+    @classmethod
+    def _get_param_names(cls):
+        pnames = set()
+        for cls_ in cls.__mro__:
+            init = cls_.__init__
+            init_signature = inspect.signature(init)
+            for p in init_signature.parameters.values():
+                if p.name == 'self':
+                    continue
+                if p.kind in {p.VAR_POSITIONAL, p.VAR_KEYWORD}:
+                    # Skip *args/**kwargs.
+                    continue
+                pnames.add(p.name)
+        return pnames
+
+    def get_params(self):
+        """Return parameters for this metric.
+
+        Returns
+        -------
+        params : dict
+            Mapping from parameter names to values.
+        """
+        return {pname: getattr(self, pname)
+                for pname in self._get_param_names()}
+
+
     def init_components(self):
         return {value: 0.0 for value in self.components_}
 
@@ -81,9 +152,6 @@ def name(self):
         """Metric name."""
         return self.metric_name()
 
-    # TODO: use joblib/locky to allow parallel processing?
-    # TODO: signature could be something like __call__(self, reference_iterator, hypothesis_iterator, ...)
-
     def __call__(self, reference: Union[Timeline, Annotation],
                  hypothesis: Union[Timeline, Annotation],
                  detailed: bool = False, uri: Optional[str] = None, **kwargs):
@@ -219,6 +287,13 @@ def __str__(self):
             sparsify=False, float_format=lambda f: "{0:.2f}".format(f)
         )
 
+    def __repr__(self):
+        cls = self.__class__.__name__
+        params = self.get_params()
+        pnames = sorted(params)
+        signature = ', '.join(f'{pname}={params[pname]}' for pname in pnames)
+        return f'{cls}({signature})'
+
     def __abs__(self):
         """Compute metric value from accumulated components"""
         return self.compute_metric(self.accumulated_)
@@ -247,6 +322,21 @@ def __iter__(self):
         for uri, component in self.results_:
             yield uri, component
 
+    def __add__(self, other):
+        cls = self.__class__
+        result = cls()
+        result.results_ = self.results_ + other.results_
+        for cname in self.components_:
+            result.accumulated_[cname] += self.accumulated_[cname]
+            result.accumulated_[cname] += other.accumulated_[cname]
+        return result
+
+    def __radd__(self, other):
+        if other == 0:
+            return self
+        else:
+            return self.__add__(other)
+
     def compute_components(self,
                            reference: Union[Timeline, Annotation],
                            hypothesis: Union[Timeline, Annotation],
@@ -320,12 +410,12 @@ def confidence_interval(self, alpha: float = 0.9) \
 
         if len(values) == 0:
             raise ValueError("Please evaluate a bunch of files before computing confidence interval.")
-        
+
         elif len(values) == 1:
             warnings.warn("Cannot compute a reliable confidence interval out of just one file.")
             center = lower = upper = values[0]
             return center, (lower, upper)
-        
+
         else:
             return scipy.stats.bayes_mvs(values, alpha=alpha)[0]
 
diff --git a/tests/test_base.py b/tests/test_base.py
new file mode 100644
index 0000000..60768a4
--- /dev/null
+++ b/tests/test_base.py
@@ -0,0 +1,157 @@
+#!/usr/bin/env python
+# encoding: utf-8
+
+# The MIT License (MIT)
+
+# Copyright (c) 2020 CNRS
+
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+
+# The above copyright notice and this permission notice shall be included in
+# all copies or substantial portions of the Software.
+
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+
+# AUTHORS
+# Hervé BREDIN - http://herve.niderb.fr
+
+
+import pytest
+
+from pyannote.core import Annotation
+from pyannote.core import Segment
+from pyannote.core import Timeline
+from pyannote.metrics.base import clone, BaseMetric
+from pyannote.metrics.detection import DetectionAccuracy
+
+
+import numpy.testing as npt
+
+# rec1
+#
+# Time        0  1  2  3  4  5  6
+# Reference   |-----|
+# Hypothesis     |-----|
+# UEM         |-----------------|
+
+# rec2
+#
+# Time        0  1  2  3  4  5  6  7
+# Reference      |--------|
+# Hypothesis           |-----|
+# UEM         |--------------------|
+
+
+@pytest.fixture
+def reference():
+    reference = {}
+    reference['rec1'] = Annotation()
+    reference['rec1'][Segment(0, 2)] = 'A'
+    reference['rec2'] = Annotation()
+    reference['rec2'][Segment(1, 4)] = 'A'
+    return reference
+
+
+@pytest.fixture
+def hypothesis():
+    hypothesis = {}
+    hypothesis['rec1'] = Annotation()
+    hypothesis['rec1'][Segment(1, 3)] = 'A'
+    hypothesis['rec2'] = Annotation()
+    hypothesis['rec2'][Segment(3, 4)] = 'A'
+    return hypothesis
+
+
+@pytest.fixture
+def uem():
+    return {
+        'rec1': Timeline([Segment(0, 6)]),
+        'rec2': Timeline([Segment(0, 7)])}
+
+
+def test_summation(reference, hypothesis, uem):
+    # Expected error rate.
+    expected = 9 / 13
+
+    # __add__
+    m1 = DetectionAccuracy()
+    m1(reference['rec1'], hypothesis['rec1'], uem=uem['rec1'])
+    m2 = DetectionAccuracy()
+    m2(reference['rec2'], hypothesis['rec2'], uem=uem['rec2'])
+    npt.assert_almost_equal(abs(m1 + m2), expected, decimal=3)
+
+    # __radd__
+    m = sum([m1, m2])
+    npt.assert_almost_equal(abs(m), expected, decimal=3)
+
+
+class M1(BaseMetric):
+    def __init__(self, a=9, **kwargs):
+        super().__init__(**kwargs)
+        self.a = a
+
+    @classmethod
+    def metric_name(cls):
+        return 'M1'
+
+    @classmethod
+    def metric_components(cls):
+        return ['c1']
+
+    def compute_metric(self, foo):
+        return 1.
+
+
+class M2(M1):
+    def __init__(self, b=10, **kwargs):
+        super().__init__(**kwargs)
+        self.b = b
+
+    @classmethod
+    def metric_name(cls):
+        return 'M2'
+
+
+def test_get_params():
+    # Subclass of BaseMetric.
+    m = M1(a=100)
+    expected = {'a': 100}
+    actual = m.get_params()
+    assert actual == expected
+
+    # Subclass of subclass of BaseMetric.
+    m = M2(a=100, b=1000)
+    expected = {'a': 100, 'b': 1000}
+    actual = m.get_params()
+    assert actual == expected
+
+
+def test_clone():
+    # Tests that clone creates deep copy of "unfit" metric.
+    metric = M1(a=10)
+    metric_new = clone(metric)
+    assert metric is not metric_new
+    assert metric.get_params() == metric_new.get_params()
+
+    # Tests that clone doesn't copy anything beyond the parameters; e.g.,
+    # results_ or accumulated_
+    metric = M1(a=10)
+    metric.accumulated_['c1'] = 999999
+    metric_new = clone(metric)
+    assert metric_new.accumulated_['c1'] == 0
+
+
+def test_repr():
+    m = M1(a=10)
+    assert repr(m) == 'M1(a=10)'