File size: 10,599 Bytes
ad16788
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
#!/usr/bin/env python3

import argparse
import logging
from pathlib import Path
import sys
from typing import List
from typing import Optional
from typing import Sequence
from typing import Tuple
from typing import Union

import numpy as np
import torch
from tqdm import trange
from typeguard import check_argument_types

from espnet.utils.cli_utils import get_commandline_args
from espnet2.fileio.npy_scp import NpyScpWriter
from espnet2.tasks.diar import DiarizationTask
from espnet2.torch_utils.device_funcs import to_device
from espnet2.torch_utils.set_all_random_seed import set_all_random_seed
from espnet2.utils import config_argparse
from espnet2.utils.types import humanfriendly_parse_size_or_none
from espnet2.utils.types import str2bool
from espnet2.utils.types import str2triple_str
from espnet2.utils.types import str_or_none


class DiarizeSpeech:
    """DiarizeSpeech class

    Examples:
        >>> import soundfile
        >>> diarization = DiarizeSpeech("diar_config.yaml", "diar.pth")
        >>> audio, rate = soundfile.read("speech.wav")
        >>> diarization(audio)
        [(spk_id, start, end), (spk_id2, start2, end2)]

    """

    def __init__(
        self,
        diar_train_config: Union[Path, str],
        diar_model_file: Union[Path, str] = None,
        segment_size: Optional[float] = None,
        normalize_segment_scale: bool = False,
        show_progressbar: bool = False,
        device: str = "cpu",
        dtype: str = "float32",
    ):
        assert check_argument_types()

        # 1. Build Diar model
        diar_model, diar_train_args = DiarizationTask.build_model_from_file(
            diar_train_config, diar_model_file, device
        )
        diar_model.to(dtype=getattr(torch, dtype)).eval()

        self.device = device
        self.dtype = dtype
        self.diar_train_args = diar_train_args
        self.diar_model = diar_model

        # only used when processing long speech, i.e.
        # segment_size is not None and hop_size is not None
        self.segment_size = segment_size
        self.normalize_segment_scale = normalize_segment_scale
        self.show_progressbar = show_progressbar

        self.num_spk = diar_model.num_spk

        self.segmenting = segment_size is not None
        if self.segmenting:
            logging.info("Perform segment-wise speaker diarization")
            logging.info("Segment length = {} sec".format(segment_size))
        else:
            logging.info("Perform direct speaker diarization on the input")

    @torch.no_grad()
    def __call__(
        self, speech: Union[torch.Tensor, np.ndarray], fs: int = 8000
    ) -> List[torch.Tensor]:
        """Inference

        Args:
            speech: Input speech data (Batch, Nsamples [, Channels])
            fs: sample rate
        Returns:
            [speaker_info1, speaker_info2, ...]

        """
        assert check_argument_types()

        # Input as audio signal
        if isinstance(speech, np.ndarray):
            speech = torch.as_tensor(speech)

        assert speech.dim() > 1, speech.size()
        batch_size = speech.size(0)
        speech = speech.to(getattr(torch, self.dtype))
        # lenghts: (B,)
        lengths = speech.new_full(
            [batch_size], dtype=torch.long, fill_value=speech.size(1)
        )

        # a. To device
        speech = to_device(speech, device=self.device)
        lengths = to_device(lengths, device=self.device)

        if self.segmenting and lengths[0] > self.segment_size * fs:
            # Segment-wise speaker diarization
            num_segments = int(np.ceil(speech.size(1) / (self.segment_size * fs)))
            t = T = int(self.segment_size * fs)
            pad_shape = speech[:, :T].shape
            diarized_wavs = []
            range_ = trange if self.show_progressbar else range
            for i in range_(num_segments):
                st = int(i * self.segment_size * fs)
                en = st + T
                if en >= lengths[0]:
                    # en - st < T (last segment)
                    en = lengths[0]
                    speech_seg = speech.new_zeros(pad_shape)
                    t = en - st
                    speech_seg[:, :t] = speech[:, st:en]
                else:
                    t = T
                    speech_seg = speech[:, st:en]  # B x T [x C]

                lengths_seg = speech.new_full(
                    [batch_size], dtype=torch.long, fill_value=T
                )
                # b. Diarization Forward
                encoder_out, encoder_out_lens = self.diar_model.encode(
                    speech_seg, lengths_seg
                )
                spk_prediction = self.diar_model.decoder(encoder_out, encoder_out_lens)

                # List[torch.Tensor(B, T, num_spks)]
                diarized_wavs.append(spk_prediction)

            spk_prediction = torch.cat(diarized_wavs, dim=1)
        else:
            # b. Diarization Forward
            encoder_out, encoder_out_lens = self.diar_model.encode(speech, lengths)
            spk_prediction = self.diar_model.decoder(encoder_out, encoder_out_lens)

        assert spk_prediction.size(2) == self.num_spk, (
            spk_prediction.size(2),
            self.num_spk,
        )
        assert spk_prediction.size(0) == batch_size, (
            spk_prediction.size(0),
            batch_size,
        )
        spk_prediction = spk_prediction.cpu().numpy()
        spk_prediction = 1 / (1 + np.exp(-spk_prediction))

        return spk_prediction


def inference(
    output_dir: str,
    batch_size: int,
    dtype: str,
    fs: int,
    ngpu: int,
    seed: int,
    num_workers: int,
    log_level: Union[int, str],
    data_path_and_name_and_type: Sequence[Tuple[str, str, str]],
    key_file: Optional[str],
    diar_train_config: str,
    diar_model_file: str,
    allow_variable_data_keys: bool,
    segment_size: Optional[float],
    show_progressbar: bool,
):
    assert check_argument_types()
    if batch_size > 1:
        raise NotImplementedError("batch decoding is not implemented")
    if ngpu > 1:
        raise NotImplementedError("only single GPU decoding is supported")

    logging.basicConfig(
        level=log_level,
        format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
    )

    if ngpu >= 1:
        device = "cuda"
    else:
        device = "cpu"

    # 1. Set random-seed
    set_all_random_seed(seed)

    # 2. Build separate_speech
    diarize_speech = DiarizeSpeech(
        diar_train_config=diar_train_config,
        diar_model_file=diar_model_file,
        segment_size=segment_size,
        show_progressbar=show_progressbar,
        device=device,
        dtype=dtype,
    )

    # 3. Build data-iterator
    loader = DiarizationTask.build_streaming_iterator(
        data_path_and_name_and_type,
        dtype=dtype,
        batch_size=batch_size,
        key_file=key_file,
        num_workers=num_workers,
        preprocess_fn=DiarizationTask.build_preprocess_fn(
            diarize_speech.diar_train_args, False
        ),
        collate_fn=DiarizationTask.build_collate_fn(
            diarize_speech.diar_train_args, False
        ),
        allow_variable_data_keys=allow_variable_data_keys,
        inference=True,
    )

    # 4. Start for-loop
    writer = NpyScpWriter(f"{output_dir}/predictions", f"{output_dir}/diarize.scp")

    for keys, batch in loader:
        assert isinstance(batch, dict), type(batch)
        assert all(isinstance(s, str) for s in keys), keys
        _bs = len(next(iter(batch.values())))
        assert len(keys) == _bs, f"{len(keys)} != {_bs}"
        batch = {k: v for k, v in batch.items() if not k.endswith("_lengths")}

        spk_predictions = diarize_speech(**batch)
        for b in range(batch_size):
            writer[keys[b]] = spk_predictions[b]

    writer.close()


def get_parser():
    parser = config_argparse.ArgumentParser(
        description="Speaker Diarization inference",
        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
    )

    # Note(kamo): Use '_' instead of '-' as separator.
    # '-' is confusing if written in yaml.
    parser.add_argument(
        "--log_level",
        type=lambda x: x.upper(),
        default="INFO",
        choices=("CRITICAL", "ERROR", "WARNING", "INFO", "DEBUG", "NOTSET"),
        help="The verbose level of logging",
    )

    parser.add_argument("--output_dir", type=str, required=True)
    parser.add_argument(
        "--ngpu",
        type=int,
        default=0,
        help="The number of gpus. 0 indicates CPU mode",
    )
    parser.add_argument("--seed", type=int, default=0, help="Random seed")
    parser.add_argument(
        "--dtype",
        default="float32",
        choices=["float16", "float32", "float64"],
        help="Data type",
    )
    parser.add_argument(
        "--fs",
        type=humanfriendly_parse_size_or_none,
        default=8000,
        help="Sampling rate",
    )
    parser.add_argument(
        "--num_workers",
        type=int,
        default=1,
        help="The number of workers used for DataLoader",
    )

    group = parser.add_argument_group("Input data related")
    group.add_argument(
        "--data_path_and_name_and_type",
        type=str2triple_str,
        required=True,
        action="append",
    )
    group.add_argument("--key_file", type=str_or_none)
    group.add_argument("--allow_variable_data_keys", type=str2bool, default=False)

    group = parser.add_argument_group("The model configuration related")
    group.add_argument("--diar_train_config", type=str, required=True)
    group.add_argument("--diar_model_file", type=str, required=True)

    group = parser.add_argument_group("Data loading related")
    group.add_argument(
        "--batch_size",
        type=int,
        default=1,
        help="The batch size for inference",
    )
    group = parser.add_argument_group("Diarize speech related")
    group.add_argument(
        "--segment_size",
        type=float,
        default=None,
        help="Segment length in seconds for segment-wise speaker diarization",
    )
    group.add_argument(
        "--show_progressbar",
        type=str2bool,
        default=False,
        help="Whether to show a progress bar when performing segment-wise speaker "
        "diarization",
    )

    return parser


def main(cmd=None):
    print(get_commandline_args(), file=sys.stderr)
    parser = get_parser()
    args = parser.parse_args(cmd)
    kwargs = vars(args)
    kwargs.pop("config", None)
    inference(**kwargs)


if __name__ == "__main__":
    main()