offline-nemo-enc-dec-ctc-model.h
2.6 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
// sherpa-onnx/csrc/offline-nemo-enc-dec-ctc-model.h
//
// Copyright (c) 2023 Xiaomi Corporation
#ifndef SHERPA_ONNX_CSRC_OFFLINE_NEMO_ENC_DEC_CTC_MODEL_H_
#define SHERPA_ONNX_CSRC_OFFLINE_NEMO_ENC_DEC_CTC_MODEL_H_
#include <memory>
#include <string>
#include <utility>
#include <vector>
#if __ANDROID_API__ >= 9
#include "android/asset_manager.h"
#include "android/asset_manager_jni.h"
#endif
#include "onnxruntime_cxx_api.h" // NOLINT
#include "sherpa-onnx/csrc/offline-ctc-model.h"
#include "sherpa-onnx/csrc/offline-model-config.h"
namespace sherpa_onnx {
/** This class implements the EncDecCTCModelBPE model from NeMo.
*
* See
* https://github.com/NVIDIA/NeMo/blob/main/nemo/collections/asr/models/ctc_bpe_models.py
* https://github.com/NVIDIA/NeMo/blob/main/nemo/collections/asr/models/ctc_models.py
*/
class OfflineNemoEncDecCtcModel : public OfflineCtcModel {
public:
explicit OfflineNemoEncDecCtcModel(const OfflineModelConfig &config);
#if __ANDROID_API__ >= 9
OfflineNemoEncDecCtcModel(AAssetManager *mgr,
const OfflineModelConfig &config);
#endif
~OfflineNemoEncDecCtcModel() override;
/** Run the forward method of the model.
*
* @param features A tensor of shape (N, T, C).
* @param features_length A 1-D tensor of shape (N,) containing number of
* valid frames in `features` before padding.
* Its dtype is int64_t.
*
* @return Return a vector containing:
* - log_probs: A 3-D tensor of shape (N, T', vocab_size).
* - log_probs_length A 1-D tensor of shape (N,). Its dtype is int64_t
*/
std::vector<Ort::Value> Forward(Ort::Value features,
Ort::Value features_length) override;
/** Return the vocabulary size of the model
*/
int32_t VocabSize() const override;
/** SubsamplingFactor of the model
*
* For Citrinet, the subsampling factor is usually 4.
* For Conformer CTC, the subsampling factor is usually 8.
*/
int32_t SubsamplingFactor() const override;
/** Return an allocator for allocating memory
*/
OrtAllocator *Allocator() const override;
// Possible values:
// - per_feature
// - all_features (not implemented yet)
// - fixed_mean (not implemented)
// - fixed_std (not implemented)
// - or just leave it to empty
// See
// https://github.com/NVIDIA/NeMo/blob/main/nemo/collections/asr/parts/preprocessing/features.py#L59
// for details
std::string FeatureNormalizationMethod() const override;
private:
class Impl;
std::unique_ptr<Impl> impl_;
};
} // namespace sherpa_onnx
#endif // SHERPA_ONNX_CSRC_OFFLINE_NEMO_ENC_DEC_CTC_MODEL_H_