PaddleSpeech/tests/deepspeech2_online_model_test.py

# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import unittest

import numpy as np
import paddle

from deepspeech.models.ds2_online import DeepSpeech2ModelOnline


class TestDeepSpeech2ModelOnline(unittest.TestCase):
    def setUp(self):
        paddle.set_device('cpu')

        self.batch_size = 2
        self.feat_dim = 161
        max_len = 210

        # (B, T, D)
        audio = np.random.randn(self.batch_size, max_len, self.feat_dim)
        audio_len = np.random.randint(max_len, size=self.batch_size)
        audio_len[-1] = max_len
        # (B, U)
        text = np.array([[1, 2], [1, 2]])
        text_len = np.array([2] * self.batch_size)

        self.audio = paddle.to_tensor(audio, dtype='float32')
        self.audio_len = paddle.to_tensor(audio_len, dtype='int64')
        self.text = paddle.to_tensor(text, dtype='int32')
        self.text_len = paddle.to_tensor(text_len, dtype='int64')

    def test_ds2_1(self):
        model = DeepSpeech2ModelOnline(
            feat_size=self.feat_dim,
            dict_size=10,
            num_conv_layers=2,
            num_rnn_layers=3,
            rnn_size=1024,
            num_fc_layers=2,
            fc_layers_size_list=[512, 256],
            use_gru=False)
        loss = model(self.audio, self.audio_len, self.text, self.text_len)
        self.assertEqual(loss.numel(), 1)

    def test_ds2_2(self):
        model = DeepSpeech2ModelOnline(
            feat_size=self.feat_dim,
            dict_size=10,
            num_conv_layers=2,
            num_rnn_layers=3,
            rnn_size=1024,
            num_fc_layers=2,
            fc_layers_size_list=[512, 256],
            use_gru=True)
        loss = model(self.audio, self.audio_len, self.text, self.text_len)
        self.assertEqual(loss.numel(), 1)

    def test_ds2_3(self):
        model = DeepSpeech2ModelOnline(
            feat_size=self.feat_dim,
            dict_size=10,
            num_conv_layers=2,
            num_rnn_layers=3,
            rnn_size=1024,
            num_fc_layers=2,
            fc_layers_size_list=[512, 256],
            use_gru=False)
        loss = model(self.audio, self.audio_len, self.text, self.text_len)
        self.assertEqual(loss.numel(), 1)

    def test_ds2_4(self):
        model = DeepSpeech2ModelOnline(
            feat_size=self.feat_dim,
            dict_size=10,
            num_conv_layers=2,
            num_rnn_layers=3,
            rnn_size=1024,
            num_fc_layers=2,
            fc_layers_size_list=[512, 256],
            use_gru=True)
        loss = model(self.audio, self.audio_len, self.text, self.text_len)
        self.assertEqual(loss.numel(), 1)

    def test_ds2_5(self):
        model = DeepSpeech2ModelOnline(
            feat_size=self.feat_dim,
            dict_size=10,
            num_conv_layers=2,
            num_rnn_layers=3,
            rnn_size=1024,
            num_fc_layers=2,
            fc_layers_size_list=[512, 256],
            use_gru=False)
        loss = model(self.audio, self.audio_len, self.text, self.text_len)
        self.assertEqual(loss.numel(), 1)

    def test_ds2_6(self):
        model = DeepSpeech2ModelOnline(
            feat_size=self.feat_dim,
            dict_size=10,
            num_conv_layers=2,
            num_rnn_layers=3,
            rnn_size=1024,
            rnn_direction='bidirect',
            num_fc_layers=2,
            fc_layers_size_list=[512, 256],
            use_gru=False)
        loss = model(self.audio, self.audio_len, self.text, self.text_len)
        self.assertEqual(loss.numel(), 1)

    def test_ds2_7(self):
        use_gru = False
        model = DeepSpeech2ModelOnline(
            feat_size=self.feat_dim,
            dict_size=10,
            num_conv_layers=2,
            num_rnn_layers=1,
            rnn_size=1024,
            rnn_direction='forward',
            num_fc_layers=2,
            fc_layers_size_list=[512, 256],
            use_gru=use_gru)
        model.eval()
        paddle.device.set_device("cpu")
        de_ch_size = 8

        eouts, eouts_lens, final_state_h_box, final_state_c_box = model.encoder(
            self.audio, self.audio_len)
        eouts_by_chk_list, eouts_lens_by_chk_list, final_state_h_box_chk, final_state_c_box_chk = model.encoder.forward_chunk_by_chunk(
            self.audio, self.audio_len, de_ch_size)
        eouts_by_chk = paddle.concat(eouts_by_chk_list, axis=1)
        eouts_lens_by_chk = paddle.add_n(eouts_lens_by_chk_list)
        decode_max_len = eouts.shape[1]
        eouts_by_chk = eouts_by_chk[:, :decode_max_len, :]
        self.assertEqual(paddle.allclose(eouts_by_chk, eouts), True)
        self.assertEqual(
            paddle.allclose(final_state_h_box, final_state_h_box_chk), True)
        if use_gru is False:
            self.assertEqual(
                paddle.allclose(final_state_c_box, final_state_c_box_chk), True)

    def test_ds2_8(self):
        use_gru = True
        model = DeepSpeech2ModelOnline(
            feat_size=self.feat_dim,
            dict_size=10,
            num_conv_layers=2,
            num_rnn_layers=1,
            rnn_size=1024,
            rnn_direction='forward',
            num_fc_layers=2,
            fc_layers_size_list=[512, 256],
            use_gru=use_gru)
        model.eval()
        paddle.device.set_device("cpu")
        de_ch_size = 8

        eouts, eouts_lens, final_state_h_box, final_state_c_box = model.encoder(
            self.audio, self.audio_len)
        eouts_by_chk_list, eouts_lens_by_chk_list, final_state_h_box_chk, final_state_c_box_chk = model.encoder.forward_chunk_by_chunk(
            self.audio, self.audio_len, de_ch_size)
        eouts_by_chk = paddle.concat(eouts_by_chk_list, axis=1)
        eouts_lens_by_chk = paddle.add_n(eouts_lens_by_chk_list)
        decode_max_len = eouts.shape[1]
        eouts_by_chk = eouts_by_chk[:, :decode_max_len, :]
        self.assertEqual(paddle.allclose(eouts_by_chk, eouts), True)
        self.assertEqual(
            paddle.allclose(final_state_h_box, final_state_h_box_chk), True)
        if use_gru is False:
            self.assertEqual(
                paddle.allclose(final_state_c_box, final_state_c_box_chk), True)


if __name__ == '__main__':
    unittest.main()