File size: 6,533 Bytes
affcd23
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import argparse
import json
import os
from pathlib import Path

import torch
from tqdm import tqdm

import hw_asr.model as module_model
from hw_asr.trainer import Trainer
from hw_asr.utils import ROOT_PATH
from hw_asr.utils.object_loading import get_dataloaders
from hw_asr.utils.parse_config import ConfigParser
from hw_asr.metric.utils import calc_wer

DEFAULT_CHECKPOINT_PATH = ROOT_PATH / "default_test_model" / "checkpoint.pth"


def main(config, out_file):
    logger = config.get_logger("test")

    # define cpu or gpu if possible
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

    # text_encoder
    text_encoder = config.get_text_encoder()

    # setup data_loader instances
    dataloaders = get_dataloaders(config, text_encoder)

    # build model architecture
    model = config.init_obj(config["arch"], module_model, n_class=len(text_encoder))
    logger.info(model)

    logger.info("Loading checkpoint: {} ...".format(config.resume))
    checkpoint = torch.load(config.resume, map_location=device)
    state_dict = checkpoint["state_dict"]
    if config["n_gpu"] > 1:
        model = torch.nn.DataParallel(model)
    model.load_state_dict(state_dict)

    # prepare model for testing
    model = model.to(device)
    model.eval()

    results = []

    argmax_wer_sum = 0
    beam_search_wer_sum = 0
    lm_wer_sum = 0

    with torch.no_grad():
        for batch_num, batch in enumerate(tqdm(dataloaders["test"])):
            batch = Trainer.move_batch_to_device(batch, device)
            output = model(**batch)
            if type(output) is dict:
                batch.update(output)
            else:
                batch["logits"] = output
            batch["log_probs"] = torch.log_softmax(batch["logits"], dim=-1)
            batch["log_probs_length"] = model.transform_input_lengths(batch["spectrogram_length"])
            batch["probs"] = batch["log_probs"].exp().cpu()
            batch["argmax"] = batch["probs"].argmax(-1)
            for i in range(len(batch["text"])):
                length = int(batch["log_probs_length"][i])
                ground_truth = batch["text"][i]

                argmax = batch["argmax"][i][:length].cpu().numpy()
                text_argmax = text_encoder.ctc_decode(argmax)

                probs = batch["probs"][i][:length].detach().cpu().numpy()
                text_beam_search = text_encoder.ctc_beam_search(probs, beam_size=4)

                logits = batch["logits"][i][:length].detach().cpu().numpy()
                text_lm = text_encoder.ctc_lm_beam_search(logits)

                argmax_wer = calc_wer(ground_truth, text_argmax) * 100
                beam_search_wer = calc_wer(ground_truth, text_beam_search) * 100
                lm_wer = calc_wer(ground_truth, text_lm) * 100

                argmax_wer_sum += argmax_wer
                beam_search_wer_sum += beam_search_wer
                lm_wer_sum += lm_wer

                results.append(
                    {
                        "ground_truth": ground_truth,
                        "pred_text_argmax": text_argmax,
                        "pred_text_beam_search": text_beam_search,
                        "pred_text_lm": text_lm,
                        "argmax_wer": argmax_wer,
                        "beam_search_wer": beam_search_wer,
                        "lm_wer": lm_wer,
                    }
                )

    n = len(results)
    logger.info("argmax_wer_mean:")
    logger.info(argmax_wer_sum / n)
    logger.info("beam_search_wer_mean:")
    logger.info(beam_search_wer_sum / n)
    logger.info("lm_wer_mean:")
    logger.info(lm_wer_sum / n)

    with Path(out_file).open("w") as f:
        json.dump(results, f, indent=2)


if __name__ == "__main__":
    args = argparse.ArgumentParser(description="PyTorch Template")
    args.add_argument(
        "-c",
        "--config",
        default=None,
        type=str,
        help="config file path (default: None)",
    )
    args.add_argument(
        "-r",
        "--resume",
        default=str(DEFAULT_CHECKPOINT_PATH.absolute().resolve()),
        type=str,
        help="path to latest checkpoint (default: None)",
    )
    args.add_argument(
        "-d",
        "--device",
        default=None,
        type=str,
        help="indices of GPUs to enable (default: all)",
    )
    args.add_argument(
        "-o",
        "--output",
        default="output.json",
        type=str,
        help="File to write results (.json)",
    )
    args.add_argument(
        "-t",
        "--test-data-folder",
        default=None,
        type=str,
        help="Path to dataset",
    )
    args.add_argument(
        "-b",
        "--batch-size",
        default=20,
        type=int,
        help="Test dataset batch size",
    )
    args.add_argument(
        "-j",
        "--jobs",
        default=1,
        type=int,
        help="Number of workers for test dataloader",
    )

    args = args.parse_args()

    # set GPUs
    if args.device is not None:
        os.environ["CUDA_VISIBLE_DEVICES"] = args.device

    # first, we need to obtain config with model parameters
    # we assume it is located with checkpoint in the same folder
    model_config = Path(args.resume).parent / "config.json"
    with model_config.open() as f:
        config = ConfigParser(json.load(f), resume=args.resume)

    # update with addition configs from `args.config` if provided
    if args.config is not None:
        with Path(args.config).open() as f:
            config.config.update(json.load(f))

    # if `--test-data-folder` was provided, set it as a default test set
    if args.test_data_folder is not None:
        test_data_folder = Path(args.test_data_folder).absolute().resolve()
        assert test_data_folder.exists()
        config.config["data"] = {
            "test": {
                "batch_size": args.batch_size,
                "num_workers": args.jobs,
                "datasets": [
                    {
                        "type": "CustomDirAudioDataset",
                        "args": {
                            "audio_dir": str(test_data_folder / "audio"),
                            "transcription_dir": str(test_data_folder / "transcriptions"),
                        },
                    }
                ],
            }
        }

    assert config.config.get("data", {}).get("test", None) is not None
    config["data"]["test"]["batch_size"] = args.batch_size
    config["data"]["test"]["n_jobs"] = args.jobs

    main(config, args.output)