offline-punctuation-ct-transformer-impl.h
5.3 KB
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
// sherpa-onnx/csrc/offline-punctuation-ct-transformer-impl.h
//
// Copyright (c) 2024 Xiaomi Corporation
#ifndef SHERPA_ONNX_CSRC_OFFLINE_PUNCTUATION_CT_TRANSFORMER_IMPL_H_
#define SHERPA_ONNX_CSRC_OFFLINE_PUNCTUATION_CT_TRANSFORMER_IMPL_H_
#include <memory>
#include <string>
#include <utility>
#include <vector>
#include "sherpa-onnx/csrc/macros.h"
#include "sherpa-onnx/csrc/math.h"
#include "sherpa-onnx/csrc/offline-ct-transformer-model.h"
#include "sherpa-onnx/csrc/offline-punctuation-impl.h"
#include "sherpa-onnx/csrc/offline-punctuation.h"
#include "sherpa-onnx/csrc/text-utils.h"
namespace sherpa_onnx {
class OfflinePunctuationCtTransformerImpl : public OfflinePunctuationImpl {
public:
explicit OfflinePunctuationCtTransformerImpl(
const OfflinePunctuationConfig &config)
: config_(config), model_(config.model) {}
std::string AddPunctuation(const std::string &text) const override {
if (text.empty()) {
return {};
}
std::vector<std::string> tokens = SplitUtf8(text);
std::vector<int32_t> token_ids;
token_ids.reserve(tokens.size());
const auto &meta_data = model_.GetModelMetadata();
for (const auto &t : tokens) {
std::string token = ToLowerCase(t);
if (meta_data.token2id.count(token)) {
token_ids.push_back(meta_data.token2id.at(token));
} else {
token_ids.push_back(meta_data.unk_id);
}
}
auto memory_info =
Ort::MemoryInfo::CreateCpu(OrtDeviceAllocator, OrtMemTypeDefault);
int32_t segment_size = 20;
int32_t max_len = 200;
int32_t num_segments = (token_ids.size() + segment_size - 1) / segment_size;
std::vector<int32_t> punctuations;
int32_t last = -1;
for (int32_t i = 0; i != num_segments; ++i) {
int32_t this_start = i * segment_size; // inclusive
int32_t this_end = this_start + segment_size; // exclusive
if (this_end > token_ids.size()) {
this_end = token_ids.size();
}
if (last != -1) {
this_start = last;
}
// token_ids[this_start:this_end] is sent to the model
std::array<int64_t, 2> x_shape = {1, this_end - this_start};
Ort::Value x =
Ort::Value::CreateTensor(memory_info, token_ids.data() + this_start,
x_shape[1], x_shape.data(), x_shape.size());
int64_t len_shape = 1;
int32_t len = x_shape[1];
Ort::Value x_len =
Ort::Value::CreateTensor(memory_info, &len, 1, &len_shape, 1);
Ort::Value out = model_.Forward(std::move(x), std::move(x_len));
// [N, T, num_punctuations]
std::vector<int64_t> out_shape =
out.GetTensorTypeAndShapeInfo().GetShape();
assert(out_shape[0] == 1);
assert(out_shape[1] == len);
assert(out_shape[2] == meta_data.num_punctuations);
std::vector<int32_t> this_punctuations;
this_punctuations.reserve(len);
const float *p = out.GetTensorData<float>();
for (int32_t k = 0; k != len; ++k, p += meta_data.num_punctuations) {
auto index = static_cast<int32_t>(std::distance(
p, std::max_element(p, p + meta_data.num_punctuations)));
this_punctuations.push_back(index);
} // for (int32_t k = 0; k != len; ++k, p += meta_data.num_punctuations)
int32_t dot_index = -1;
int32_t comma_index = -1;
for (int32_t m = this_punctuations.size() - 1; m >= 1; --m) {
int32_t punct_id = this_punctuations[m];
if (punct_id == meta_data.dot_id || punct_id == meta_data.quest_id) {
dot_index = m;
break;
}
if (comma_index == -1 && punct_id == meta_data.comma_id) {
comma_index = m;
}
} // for (int32_t k = this_punctuations.size() - 1; k >= 1; --k)
if (dot_index == -1 && len >= max_len && comma_index != -1) {
dot_index = comma_index;
this_punctuations[dot_index] = meta_data.dot_id;
}
if (dot_index == -1) {
if (last == -1) {
last = this_start;
}
if (i == num_segments - 1) {
dot_index = token_ids.size() - 1;
}
} else {
last = this_start + dot_index + 1;
punctuations.insert(punctuations.end(), this_punctuations.begin(),
this_punctuations.begin() + (dot_index + 1));
}
} // for (int32_t i = 0; i != num_segments; ++i)
if (punctuations.size() != token_ids.size() &&
punctuations.size() + 1 == token_ids.size()) {
punctuations.push_back(meta_data.dot_id);
}
if (punctuations.size() != token_ids.size()) {
SHERPA_ONNX_LOGE("%s, %d, %d. Some unexpected things happened",
text.c_str(), static_cast<int32_t>(punctuations.size()),
static_cast<int32_t>(token_ids.size()));
return text;
}
std::string ans;
for (int32_t i = 0; i != static_cast<int32_t>(punctuations.size()); ++i) {
const std::string &w = tokens[i];
if (i > 0 && !(ans.back() & 0x80) && !(w[0] & 0x80)) {
ans.push_back(' ');
}
ans.append(w);
if (punctuations[i] != meta_data.underline_id) {
ans.append(meta_data.id2punct[punctuations[i]]);
}
}
return ans;
}
private:
OfflinePunctuationConfig config_;
OfflineCtTransformerModel model_;
};
} // namespace sherpa_onnx
#endif // SHERPA_ONNX_CSRC_OFFLINE_PUNCTUATION_CT_TRANSFORMER_IMPL_H_