online-lstm-transducer-model.h
3.0 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
// sherpa-onnx/csrc/online-lstm-transducer-model.h
//
// Copyright (c) 2023 Xiaomi Corporation
#ifndef SHERPA_ONNX_CSRC_ONLINE_LSTM_TRANSDUCER_MODEL_H_
#define SHERPA_ONNX_CSRC_ONLINE_LSTM_TRANSDUCER_MODEL_H_
#include <memory>
#include <string>
#include <utility>
#include <vector>
#include "onnxruntime_cxx_api.h" // NOLINT
#include "sherpa-onnx/csrc/online-model-config.h"
#include "sherpa-onnx/csrc/online-transducer-model.h"
namespace sherpa_onnx {
class OnlineLstmTransducerModel : public OnlineTransducerModel {
public:
explicit OnlineLstmTransducerModel(const OnlineModelConfig &config);
template <typename Manager>
OnlineLstmTransducerModel(Manager *mgr, const OnlineModelConfig &config);
std::vector<Ort::Value> StackStates(
const std::vector<std::vector<Ort::Value>> &states) const override;
std::vector<std::vector<Ort::Value>> UnStackStates(
const std::vector<Ort::Value> &states) const override;
std::vector<Ort::Value> GetEncoderInitStates() override;
std::pair<Ort::Value, std::vector<Ort::Value>> RunEncoder(
Ort::Value features, std::vector<Ort::Value> states,
Ort::Value processed_frames) override;
Ort::Value RunDecoder(Ort::Value decoder_input) override;
Ort::Value RunJoiner(Ort::Value encoder_out, Ort::Value decoder_out) override;
int32_t ContextSize() const override { return context_size_; }
int32_t ChunkSize() const override { return T_; }
int32_t ChunkShift() const override { return decode_chunk_len_; }
int32_t VocabSize() const override { return vocab_size_; }
OrtAllocator *Allocator() override { return allocator_; }
private:
void InitEncoder(void *model_data, size_t model_data_length);
void InitDecoder(void *model_data, size_t model_data_length);
void InitJoiner(void *model_data, size_t model_data_length);
private:
Ort::Env env_;
Ort::SessionOptions sess_opts_;
Ort::AllocatorWithDefaultOptions allocator_;
std::unique_ptr<Ort::Session> encoder_sess_;
std::unique_ptr<Ort::Session> decoder_sess_;
std::unique_ptr<Ort::Session> joiner_sess_;
std::vector<std::string> encoder_input_names_;
std::vector<const char *> encoder_input_names_ptr_;
std::vector<std::string> encoder_output_names_;
std::vector<const char *> encoder_output_names_ptr_;
std::vector<std::string> decoder_input_names_;
std::vector<const char *> decoder_input_names_ptr_;
std::vector<std::string> decoder_output_names_;
std::vector<const char *> decoder_output_names_ptr_;
std::vector<std::string> joiner_input_names_;
std::vector<const char *> joiner_input_names_ptr_;
std::vector<std::string> joiner_output_names_;
std::vector<const char *> joiner_output_names_ptr_;
OnlineModelConfig config_;
int32_t num_encoder_layers_ = 0;
int32_t T_ = 0;
int32_t decode_chunk_len_ = 0;
int32_t rnn_hidden_size_ = 0;
int32_t d_model_ = 0;
int32_t context_size_ = 0;
int32_t vocab_size_ = 0;
};
} // namespace sherpa_onnx
#endif // SHERPA_ONNX_CSRC_ONLINE_LSTM_TRANSDUCER_MODEL_H_