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
|
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"lm_weights = [0, 1.0, 2.5,]\n",
"word_score = [-1.5, 0.0, 1.5]\n",
"beam_sizes = [50, 500]\n",
"beam_thresholds = [50]\n",
"beam_size_token = [10, 38]"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from tqdm.autonotebook import tqdm\n",
"\n",
"import torch\n",
"from torch.utils.data import DataLoader\n",
"import torch.nn.functional as F\n",
"\n",
"from swr2_asr.utils.decoder import decoder_factory\n",
"from swr2_asr.utils.tokenizer import CharTokenizer\n",
"from swr2_asr.model_deep_speech import SpeechRecognitionModel\n",
"from swr2_asr.utils.data import MLSDataset, Split, DataProcessing\n",
"from swr2_asr.utils.loss_scores import cer, wer"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"\n",
"\n",
"tokenizer = CharTokenizer.from_file(\"data/tokenizers/char_tokenizer_german.json\")\n",
"\n",
"# manually increment tqdm progress bar\n",
"pbar = tqdm(total=len(lm_weights) * len(word_score) * len(beam_sizes) * len(beam_thresholds) * len(beam_size_token))\n",
"\n",
"base_config = {\n",
" \"language\": \"german\",\n",
" \"language_model_path\": \"data\", # path where model and supplementary files are stored\n",
" \"n_gram\": 3, # n-gram size of ,the language model, 3 or 5\n",
" \"beam_size\": 50 ,\n",
" \"beam_threshold\": 50,\n",
" \"n_best\": 1,\n",
" \"lm_weight\": 2,\n",
" \"word_score\": 0,\n",
" }\n",
"\n",
"dataset_params = {\n",
" \"dataset_path\": \"/Volumes/pherkel 2/SWR2-ASR\",\n",
" \"language\": \"mls_german_opus\",\n",
" \"split\": Split.DEV,\n",
" \"limited\": True,\n",
" \"download\": False,\n",
" \"size\": 0.01,\n",
"}\n",
" \n",
"\n",
"model_params = {\n",
" \"n_cnn_layers\": 3,\n",
" \"n_rnn_layers\": 5,\n",
" \"rnn_dim\": 512,\n",
" \"n_class\": tokenizer.get_vocab_size(),\n",
" \"n_feats\": 128,\n",
" \"stride\": 2,\n",
" \"dropout\": 0.1,\n",
"}\n",
"\n",
"model = SpeechRecognitionModel(**model_params)\n",
"\n",
"checkpoint = torch.load(\"data/epoch67\", map_location=torch.device(\"cpu\"))\n",
"\n",
"state_dict = {\n",
" k[len(\"module.\") :] if k.startswith(\"module.\") else k: v\n",
" for k, v in checkpoint[\"model_state_dict\"].items()\n",
"}\n",
"model.load_state_dict(state_dict, strict=True)\n",
"model.eval()\n",
"\n",
"\n",
"dataset = MLSDataset(**dataset_params,)\n",
"\n",
"data_processing = DataProcessing(\"valid\", tokenizer, {\"n_feats\": model_params[\"n_feats\"]})\n",
"\n",
"dataloader = DataLoader(\n",
" dataset=dataset,\n",
" batch_size=16,\n",
" shuffle = False,\n",
" collate_fn=data_processing,\n",
" num_workers=8,\n",
" pin_memory=True,\n",
")\n",
"\n",
"best_wer = 1.0\n",
"best_cer = 1.0\n",
"best_config = None\n",
"\n",
"for lm_weight in lm_weights:\n",
" for ws in word_score:\n",
" for beam_size in beam_sizes:\n",
" for beam_threshold in beam_thresholds:\n",
" for beam_size_t in beam_size_token:\n",
" config = base_config.copy()\n",
" config[\"lm_weight\"] = lm_weight\n",
" config[\"word_score\"] = ws\n",
" config[\"beam_size\"] = beam_size\n",
" config[\"beam_threshold\"] = beam_threshold\n",
" config[\"beam_size_token\"] = beam_size_t\n",
" \n",
" decoder = decoder_factory(\"lm\")(tokenizer, {\"lm\": config})\n",
" \n",
" test_cer, test_wer = [], []\n",
" with torch.no_grad():\n",
" model.eval()\n",
" for batch in dataloader:\n",
" # perform inference, decode, compute WER and CER\n",
" spectrograms, labels, input_lengths, label_lengths = batch\n",
" \n",
" output = model(spectrograms)\n",
" output = F.log_softmax(output, dim=2)\n",
" \n",
" decoded_preds = decoder(output)\n",
" decoded_targets = tokenizer.decode_batch(labels)\n",
" \n",
" for j, _ in enumerate(decoded_preds):\n",
" if j >= len(decoded_targets):\n",
" break\n",
" pred = \" \".join(decoded_preds[j][0].words).strip()\n",
" target = decoded_targets[j]\n",
" \n",
" test_cer.append(cer(pred, target))\n",
" test_wer.append(wer(pred, target))\n",
"\n",
" avg_cer = sum(test_cer) / len(test_cer)\n",
" avg_wer = sum(test_wer) / len(test_wer)\n",
" \n",
" if avg_wer < best_wer:\n",
" best_wer = avg_wer\n",
" best_cer = avg_cer\n",
" best_config = config\n",
" print(\"New best WER: \", best_wer, \" CER: \", best_cer)\n",
" print(\"Config: \", best_config)\n",
" print(\"LM Weight: \", lm_weight, \n",
" \" Word Score: \", ws, \n",
" \" Beam Size: \", beam_size, \n",
" \" Beam Threshold: \", beam_threshold, \n",
" \" Beam Size Token: \", beam_size_t)\n",
" print(\"--------------------------------------------------------------\")\n",
" \n",
" pbar.update(1)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": ".venv",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.13"
},
"orig_nbformat": 4
},
"nbformat": 4,
"nbformat_minor": 2
}
|
