Merge branch 'update_avspoof' into 'master'

Version 2 of AVspoof database

Created files for version 2 of AVspoof database. The new set of protocols was added that allow to use train set in the verification evaluations.

See merge request !6

advanced/databases/avspoof/2.py

+###############################################################################
+#                                                                             #
+# Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/           #
+# Contact: beat.support@idiap.ch                                              #
+#                                                                             #
+# This file is part of the beat.examples module of the BEAT platform.         #
+#                                                                             #
+# Commercial License Usage                                                    #
+# Licensees holding valid commercial BEAT licenses may use this file in       #
+# accordance with the terms contained in a written agreement between you      #
+# and Idiap. For further information contact tto@idiap.ch                     #
+#                                                                             #
+# Alternatively, this file may be used under the terms of the GNU Affero      #
+# Public License version 3 as published by the Free Software and appearing    #
+# in the file LICENSE.AGPL included in the packaging of this file.            #
+# The BEAT platform is distributed in the hope that it will be useful, but    #
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY  #
+# or FITNESS FOR A PARTICULAR PURPOSE.                                        #
+#                                                                             #
+# You should have received a copy of the GNU Affero Public License along      #
+# with the BEAT platform. If not, see http://www.gnu.org/licenses/.           #
+#                                                                             #
+###############################################################################
+
+import numpy
+import bob.io.base
+import bob.db.avspoof
+import scipy.io.wavfile
+
+from bob.db.avspoof.driver import Interface
+INFO = Interface()
+SQLITE_FILE = INFO.files()[0]
+
+
+class RecognitionTraining:
+
+    def setup(self, root_folder, outputs, parameters, force_start_index=None, force_end_index=None):
+        self.root_folder = root_folder
+        self.outputs     = outputs
+        self.parameters  = parameters
+
+        # verification.Database implements bob.db.verification.utils interface
+        self.db = bob.db.avspoof.verification.Database(original_directory=self.root_folder, original_extension='.wav')
+
+        # retrieve all file paths for the given parameters
+        self.objs = sorted(self.db.objects(protocol=parameters['protocol'], groups=parameters['group']), key=lambda x: x.id)
+
+        self.next_index       = 0
+
+        self.force_start_index = force_start_index
+        self.force_end_index = force_end_index
+
+        # Retrieve only 'useful' data, as ignore everything before force_start_index and after force_end_index
+        ### Start index
+        if self.force_start_index is not None:
+            self.next_index = self.force_start_index
+        ### End index
+        if self.force_end_index is not None:
+            self.objs = self.objs[:self.force_end_index+1]
+
+        return True
+
+
+    def done(self):
+        return (self.next_index >= len(self.objs))
+
+
+    def next(self):
+
+        obj = self.objs[self.next_index]
+
+        if self.outputs['file_id'].isConnected():
+            self.outputs['file_id'].write({'text': obj.id}, self.next_index)
+
+        if self.outputs['client_id'].isConnected():
+            self.outputs['client_id'].write({'text': obj.client_id}, self.next_index)
+
+        # read audio data
+        if self.outputs['speech'].isConnected():
+            rate, audio = scipy.io.wavfile.read(obj.make_path(self.root_folder, self.db.original_extension))
+            data = {
+                'value': numpy.cast['float'](audio)
+            }
+            self.outputs['speech'].write(data, self.next_index)
+
+        self.next_index += 1
+
+        return True
+
+
+class RecognitionTemplates:
+    # Reasoning: Each client may have a number of models in certain databases.
+    # So, each model receives an unique identifier. Those identifiers are linked
+    # to the client identifier and contain a number of samples to generate the
+    # model from.
+
+    def setup(self, root_folder, outputs, parameters, force_start_index=None, force_end_index=None):
+        self.root_folder            = root_folder
+        self.outputs                = outputs
+        self.parameters             = parameters
+
+        # verification.Database implements bob.db.verification.utils interface
+        self.db                     = bob.db.avspoof.verification.Database(original_directory=self.root_folder, original_extension='.wav')
+
+        # template Ids are model Ids (which again are just Ids of clients) in the terminoloy of the bob.db interfaces
+        self.template_ids           = sorted(self.db.model_ids(groups=parameters['group']))
+        self.objs = []
+        # store the File objects for each template
+        for template_id in self.template_ids:
+          self.objs.append(sorted(self.db.objects(protocol=parameters['protocol'], purposes=parameters['purpose'], groups=parameters['group'], model_ids = (template_id,)), key=lambda x: x.id))
+
+        self.current_template_index = 0
+        self.current_obj_index      = 0
+        self.next_index             = 0
+
+        self.force_start_index = force_start_index
+        self.force_end_index   = force_end_index
+
+        # Retrieve only 'useful' data
+        ### End index
+        #if self.force_end_index is not None:
+        #    self.objs = self.objs[:self.force_end_index+1]
+        ### Start index
+        if self.force_start_index is None:
+            self.force_start_index = 0
+        while self.next_index < self.force_start_index:
+            self.next()
+
+        return True
+
+
+    def done(self):
+        return (self.current_template_index >= len(self.template_ids)) or (self.force_end_index is not None and self.force_end_index < self.next_index)
+
+
+    def next(self):
+        # the length of the batch of File objects in the current template
+        len_template_objs = len(self.objs[self.current_template_index])
+        # current file object that we want to read and output
+        obj = self.objs[self.current_template_index][self.current_obj_index]
+
+        if self.current_obj_index == 0:
+            # we update the 'template_id' and 'client_id' only when the current template/model changes
+            if self.force_start_index <= self.next_index and (self.force_end_index is None or self.force_end_index >= self.next_index):
+                # For this database, 'self.template_ids[self.current_template_index]' corresponds to 'obj.client_id'
+                if self.outputs['template_id'].isConnected():
+                    self.outputs['template_id'].write({'text': self.template_ids[self.current_template_index]}, self.next_index+len_template_objs-1)
+
+                if self.outputs['client_id'].isConnected():
+                    self.outputs['client_id'].write({'text': obj.client_id}, self.next_index+len_template_objs-1)
+
+        if self.outputs['file_id'].isConnected() or self.outputs['speech'].isConnected():
+            if self.force_start_index <= self.next_index and (self.force_end_index is None or self.force_end_index >= self.next_index):
+                if self.outputs['file_id'].isConnected():
+                    self.outputs['file_id'].write({'text': obj.id}, self.next_index)
+
+                if self.outputs['speech'].isConnected():
+                    rate, audio = scipy.io.wavfile.read(obj.make_path(self.root_folder, self.db.original_extension))
+                    data = {
+                        'value': numpy.cast['float'](audio)
+                    }
+                    self.outputs['speech'].write(data, self.next_index)
+
+            self.next_index += 1
+            self.current_obj_index += 1
+        else:
+            self.next_index += len_template_objs
+            self.current_obj_index = len_template_objs
+
+        if self.current_obj_index == len_template_objs:
+            self.current_obj_index = 0
+            self.current_template_index += 1
+
+        return True
+
+
+class ProbesReal:
+    # Reasoning: Each client may have a number of probes. Each probe may be
+    # composed of any number of sample. So, each probe receives an unique
+    # identifier. Those identifiers are linked to the client identifier and
+    # contain a number of images to generated the match/comparison score. Each
+    # probe must be matched against a number of models defined by a list of
+    # client identifiers.
+
+    def setup(self, root_folder, outputs, parameters, force_start_index=None, force_end_index=None):
+        self.root_folder = root_folder
+        self.outputs     = outputs
+        self.parameters  = parameters
+
+        # verification.Database implements bob.db.verification.utils interface
+        self.db                     = bob.db.avspoof.verification.Database(original_directory=self.root_folder, original_extension='.wav')
+
+        # template Ids are model Ids (which again are just Ids of clients) in the terminoloy of the bob.db interfaces
+        self.template_ids           = sorted(self.db.model_ids(groups=parameters['group']))
+        self.objs                   = sorted(self.db.objects(protocol=parameters['protocol'],
+                                                      purposes=parameters['purpose'], groups=parameters['group']), key=lambda x: x.id)
+
+        self.next_index  = 0
+
+        self.force_start_index = force_start_index
+        self.force_end_index = force_end_index
+
+        # Retrieve only 'useful' data
+        ### Start index
+        if self.force_start_index is not None:
+            self.next_index = self.force_start_index
+        ### End index
+        if self.force_end_index is not None:
+            if self.force_end_index < len(self.objs):
+                self.objs = self.objs[:self.force_end_index+1]
+
+        return True
+
+
+    def done(self):
+        return (self.next_index >= len(self.objs))
+
+
+    def next(self):
+
+        obj = self.objs[self.next_index]
+
+        if self.outputs['file_id'].isConnected():
+            self.outputs['file_id'].write({'text': obj.id}, self.next_index)
+
+        if self.outputs['probe_id'].isConnected():
+            self.outputs['probe_id'].write({'text': obj.id}, self.next_index)
+
+        if self.outputs['client_id'].isConnected():
+            self.outputs['client_id'].write({'text': obj.client_id}, self.next_index)
+
+        if self.outputs['template_ids'].isConnected():
+              data = {
+                  'text': numpy.array(self.template_ids, dtype=numpy.string_)
+              }
+              self.outputs['template_ids'].write(data, self.next_index)
+
+        # read audio data
+        if self.outputs['speech'].isConnected():
+            rate, audio = scipy.io.wavfile.read(obj.make_path(self.root_folder, self.db.original_extension))
+            data = {
+                'value': numpy.cast['float'](audio)
+            }
+            self.outputs['speech'].write(data, self.next_index)
+
+        self.next_index += 1
+
+        return True
+
+
+class SimpleAntispoofing:
+
+    def setup(self, root_folder, outputs, parameters, force_start_index=None, force_end_index=None):
+        self.root_folder = root_folder
+        self.outputs     = outputs
+        self.parameters  = parameters
+
+        # By right we should take the bob.db.avspoof.spoofing.Database(), which implements antispoofing.utils.db interface, but that interface does not support Id's for file objects (see antispoofing.utils.File). It's a problem for us here, since we need to sort files by their Id's and also we need to yeild the id of each file.
+        # So, we are using the direct access to the Database here.
+        self.db = bob.db.avspoof.Database()
+
+        # retrieve all file paths for the given parameters
+        self.objs_real = sorted(self.db.objects(protocol=parameters['protocol'], groups=parameters['group'], cls='real'), key=lambda x: x.id)
+        self.objs_attack = sorted(self.db.objects(protocol=parameters['protocol'], groups=parameters['group'], cls='attack'), key=lambda x: x.id)
+
+        # we have two sets of data, so to serve them sequentially, we need to know their sizes and
+        # when the second set is following after the first set, hence, this approach with offsets
+        self.offset_real    = 0
+        self.offset_attack      = self.offset_real + len(self.objs_real)
+        self.next_index       = 0
+
+        self.force_start_index = force_start_index
+        self.force_end_index = force_end_index
+
+        # Retrieve only 'useful' data, as ignore everything before force_start_index and after force_end_index
+        ### Start index
+        if self.force_start_index is not None:
+            self.next_index = self.force_start_index
+        ### End index
+        if self.force_end_index is not None:
+            if self.force_end_index < self.offset_attack:  # it means we are still within real objects
+                self.objs_real = self.objs_real[:self.force_end_index+1]
+                self.offset_attack = []
+            elif self.force_end_index < self.offset_attack + len(self.objs_attack) - 1:
+                self.objs_attack = self.objs_attack[:self.force_end_index+1-self.offset_attack]
+
+        return True
+
+
+    def done(self):
+        return (self.next_index >= len(self.objs_real) + len(self.objs_attack))
+
+
+    def next(self):
+
+        cls = ''
+        attack_type = ''
+        obj = None
+        # if we are inside set of attacks, we yeild an attack object
+        if self.next_index >= self.offset_attack:
+          obj = self.objs_attack[self.next_index - self.offset_attack]
+          cls = 'attack'
+          # get the attack and compute the type of the attack
+          attack = obj.get_attack()
+          attack_type = attack.attack_support + attack.attack_device
+        else:
+          obj = self.objs_real[self.next_index - self.offset_real]
+          cls = 'real'
+          attack_type = 'human'
+
+        if self.outputs['file_id'].isConnected():
+            self.outputs['file_id'].write({'text': obj.id}, self.next_index)
+
+        if self.outputs['client_id'].isConnected():
+            self.outputs['client_id'].write({'text': obj.client_id}, self.next_index)
+
+        # read audio data
+        if self.outputs['speech'].isConnected():
+            rate, audio = scipy.io.wavfile.read(obj.make_path(directory=self.root_folder, extension='.wav'))
+            data = {
+                'value': numpy.cast['float'](audio)
+            }
+            self.outputs['speech'].write(data, self.next_index)
+
+        if self.outputs['attack_type'].isConnected():
+            self.outputs['attack_type'].write({'text': attack_type}, self.next_index)
+
+        if self.outputs['class'].isConnected():
+            self.outputs['class'].write({'text': cls}, self.next_index)
+
+        self.next_index += 1
+
+        return True
+

advanced/databases/avspoof/2.rst

+.. Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/          ..
+.. Contact: beat.support@idiap.ch                                             ..
+..                                                                            ..
+.. This file is part of the beat.examples module of the BEAT platform.        ..
+..                                                                            ..
+.. Commercial License Usage                                                   ..
+.. Licensees holding valid commercial BEAT licenses may use this file in      ..
+.. accordance with the terms contained in a written agreement between you     ..
+.. and Idiap. For further information contact tto@idiap.ch                    ..
+..                                                                            ..
+.. Alternatively, this file may be used under the terms of the GNU Affero     ..
+.. Public License version 3 as published by the Free Software and appearing   ..
+.. in the file LICENSE.AGPL included in the packaging of this file.           ..
+.. The BEAT platform is distributed in the hope that it will be useful, but   ..
+.. WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY ..
+.. or FITNESS FOR A PARTICULAR PURPOSE.                                       ..
+..                                                                            ..
+.. You should have received a copy of the GNU Affero Public License along     ..
+.. with the BEAT platform. If not, see http://www.gnu.org/licenses/.          ..
+
+
+The AVspoof Database
+--------------------
+
+Changelog
+=========
+
+* **Version 2**, 24/Mar/2016:
+
+  - Added ``_verify_train`` protocols that allow using train set in the 
+    verification evaluations.
+
+* **Version 1**, 09/Feb/2016:
+
+  - Initial release
+
+
+Description
+===========
+
+The `AVspoof Database <https://www.idiap.ch/dataset/avspoof>`_ provides
+non-biased spoofing attacks in order for researchers to test both their
+ASV systems and anti-spoofing algorithms. The attacks are created based on
+newly acquired audio recordings. The data acquisition process lasted approximately
+two months with 44 persons, each participating in several sessions configured in
+different environmental conditions and setups. After the collection of the data,
+the attacks, more precisely, replay, voice conversion and speech synthesis attacks
+were generated. This Database was produced at the Idiap Research Institute, in Switzerland.
+
+
+Acknowledgements
+================
+
+If you use this database, please cite the following publication on your paper::
+
+  @INPROCEEDINGS{KucurErgunay_IEEEBTAS_2015,
+           author = {Kucur Ergunay, Serife and Khoury, Elie and Lazaridis, Alexandros and Marcel, S{\'{e}}bastien},
+         projects = {Idiap, SNSF-LOBI, BEAT},
+            month = sep,
+            title = {On the Vulnerability of Speaker Verification to Realistic Voice Spoofing},
+        booktitle = {IEEE International Conference on Biometrics: Theory, Applications and Systems},
+             year = {2015},
+              pdf = {http://publications.idiap.ch/downloads/papers/2015/KucurErgunay_IEEEBTAS_2015.pdf}
+  }
+
+
+Database Description
+====================
+
+The data acquisition process is divided into four different sessions, each scheduled several
+days apart in different setups and environmental conditions (e.g. different in terms of
+background noise, reverberation, etc.) for each of 31 male and 13 female participants.
+The first session which is supposed to be used as training set while creating the attacks,
+was performed in the most controlled conditions. Besides, the conditions for the last
+three sessions dedicated to test trials were more relaxed in order to grasp the challenging
+scenarios. The audio data were recorded by three different devices including (a) one
+good-quality microphone, AT2020USB+, and two mobiles, (b) Samsung Galaxy S4 (phone1)
+and (c) iPhone 3GS (phone2). The positioning of the devices was stabilized for each
+session and each participant in order to standardize the recording settings.
+
+For each session, the participant was subjected to three different data acquisition protocols as in the following:
+* **Reading part (read)**: 10/40 pre-defined sentences are read by the participant.
+* **Pass-phrases part (pass)**: 5 short prompts are read by the participant.
+* **Free speech part (free)**: The participant speaks freely about any topic for 3 to 10 minutes.
+
+The number, the length, as well as the content of the sentences for the reading and
+pass-phrases part are carefully selected in order to satisfy the constraints in terms
+of readability, data acquisition and attack quality. Similarly, the minimum duration of
+the free speech part is also determined according to our preliminary investigations
+mostly on the voice conversion attacks for which the free speech data would be included
+in the training set.
+
+Spoofing Attacks
+================
+
+In the spoofing attack creation phase, we considered creating spoofing trials for the
+text-dependent utterances of the testing data, i.e. reading parts of sessions 2-4
+and the pass-phrases of all four sessions. As a preliminary step before the creation
+of the attacks, the speech data originally recorded at 44.1 KHz sampling rate is
+down-sampled to 16 KHz.
+
+There are four main spoofing attacks for ASV systems: Impersonation, replay,
+speech synthesis and voice conversion. As the impersonation is known not to be a
+serious threat for ASV systems, we did not include it in our database. For the
+remaining three spoofing types, we designed ten different scenarios (see table below).
+We gave special attention to physical access attacks. These attacks are more realistic
+than logic access attacks considering the fact that the attacker often has no direct
+access to the system. The acquisition devices (sensors) are open to anyone, therefore
+more subjected to such attacks.
+
+
++-----------------------+--------------+-------------+--------------+-------------+
+|                       | Num. of trials per speaker |     Total num. of trials   |
++       Attacks         +--------------+-------------+--------------+-------------+
+|                       |     Male     |    Female   |     Male     |    Female   |
++-----------------------+--------------+-------------+--------------+-------------+
+|Replay-phone1          |      50      |      50     |     1550     |     650     |
++-----------------------+--------------+-------------+--------------+-------------+
+|Replay-phone2          |      50      |      50     |     1550     |     650     |
++-----------------------+--------------+-------------+--------------+-------------+
+|Replay-laptop          |      50      |      50     |     1550     |     650     |
++-----------------------+--------------+-------------+--------------+-------------+
+|Replay-laptop-HQ       |      50      |      50     |     1550     |     650     |
++-----------------------+--------------+-------------+--------------+-------------+
+|Speech-Synthesis-LA    |      35      |      35     |     1085     |     455     |
++-----------------------+--------------+-------------+--------------+-------------+
+|Speech-Synthesis-PA    |      35      |      35     |     1085     |     455     |
++-----------------------+--------------+-------------+--------------+-------------+
+|Speech-Synthesis-PA-HQ |      35      |      35     |     1085     |     455     |
++-----------------------+--------------+-------------+--------------+-------------+
+|Voice-Conversion-LA    |    1500      |     600     |    46500     |    7800     |
++-----------------------+--------------+-------------+--------------+-------------+
+|Voice-Conversion-PA    |    1500      |     600     |    46500     |    7800     |
++-----------------------+--------------+-------------+--------------+-------------+
+|Voice-Conversion-PA-HQ |    1500      |     600     |    46500     |    7800     |
++-----------------------+--------------+-------------+--------------+-------------+
+
+
+Replay Attacks
+--------------
+
+A replay attack consists of replaying a pre-recorded speech to an ASV system.
+We assume that the ASV system has a good quality microphone and the replay attack targets this sensor:
+
+* **Replay-phone1**: Replay attack using the data captured by the Samsung mobile. The speech recorded by this mobile is replayed using its own speakers and re-recorded by the microphone of the ASV system.
+* **Replay-phone2**: Replay attack using the data captured by the iPhone mobile. The speech recorded by this mobile is replayed using its own speakers and re-recorded by the microphone of the ASV system.
+* **Replay-laptop**: Replay attack using the data captured by the microphone of the ASV system. The speech recorded by this microphone is replayed using the laptop speakers and re-recorded again by the microphone of the system.
+* **Replay-laptop-HQ**: Replay attack using the data captured by the microphone of the ASV system. The speech recorded by this microphone is replayed using external high-quality loudspeakers and re-recorded using the microphone of the ASV system.
+
+Speech Synthesis Attacks
+------------------------
+
+The speech synthesis attacks were based on statistical parametric speech synthesis (SPSS).
+More specific, hidden Markov model (HMM)-based speech synthesis technique was used.
+
+* **Speech-Synthesis-LA**: Speech synthesis via logical access. The synthesized speech is directly presented to the ASV system without being re-recorded.
+* **Speech-Synthesis-PA**: Speech synthesis via physical access. The synthesized speech is replayed using the laptop speakers and re-recorded by the microphone of the ASV system.
+* **Speech-Synthesis-PA-HQ**: Speech synthesis via high-quality physical access. The synthesized speech is replayed using external high-quality loudspeakers and re-recorded by the microphone of the ASV system.
+
+Voice Conversion Attacks
+------------------------
+
+The voice conversion attacks were created using Festvox. A conversion function
+for each pair of source-target speaker is found based on the learned GMM model/parameters
+by using the source and target speakers training data. We did not consider cross-gender
+voice conversion attacks, that is only male-to-male and female-to-female conversions
+were taken into account. As in the case of speech synthesis, three possible scenarios are involved:
+
+* **Voice-Conversion-LA**: Voice conversion via logical access. The converted speech is directly presented to the system without being re-recorded.
+* **Voice-Conversion-PA**: Voice conversion via physical access. The converted speech is replayed using the speakers of the laptop and re-recorded by the microphone of the ASV system.
+* **Voice-Conversion-PA-HQ**: Voice conversion via high-quality physical access. The converted speech is replayed using external high-quality loudspeakers and re-recorded by the microphone of the ASV system.
+
+Specificities to the BEAT View
+==============================
+
+Spoofing and genuine samples are each labelled with a text field that defines
+the class of the sample: ``"attack"`` or ``"real"`` for simple anti-spoofing
+binary classification systems. Code using this
+database views may use the ``class`` field to differentiate samples.
+
+The view supports the following protocols, which are also available in the database: 
+``smalltest`` (for prove of concept experiments only, as a subset of only three clients is 
+provided for each set), ``grandtest`` (data of the whole database is provided), ``physical_access`` 
+(only replay/presentation attack are provided), and ``logical_access`` (only logical access 
+attacks are provided with no replay attacks).