online-stream.h
3.8 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
// sherpa-onnx/csrc/online-stream.h
//
// Copyright (c) 2023 Xiaomi Corporation
#ifndef SHERPA_ONNX_CSRC_ONLINE_STREAM_H_
#define SHERPA_ONNX_CSRC_ONLINE_STREAM_H_
#include <memory>
#include <vector>
#include "kaldi-decoder/csrc/faster-decoder.h"
#include "onnxruntime_cxx_api.h" // NOLINT
#include "sherpa-onnx/csrc/context-graph.h"
#include "sherpa-onnx/csrc/features.h"
#include "sherpa-onnx/csrc/online-ctc-decoder.h"
#include "sherpa-onnx/csrc/online-paraformer-decoder.h"
#include "sherpa-onnx/csrc/online-transducer-decoder.h"
namespace sherpa_onnx {
struct TransducerKeywordResult;
class OnlineStream {
public:
explicit OnlineStream(const FeatureExtractorConfig &config = {},
ContextGraphPtr context_graph = nullptr);
~OnlineStream();
/**
@param sampling_rate The sampling_rate of the input waveform. If it does
not equal to config.sampling_rate, we will do
resampling inside.
@param waveform Pointer to a 1-D array of size n. It must be normalized to
the range [-1, 1].
@param n Number of entries in waveform
*/
void AcceptWaveform(int32_t sampling_rate, const float *waveform,
int32_t n) const;
/**
* InputFinished() tells the class you won't be providing any
* more waveform. This will help flush out the last frame or two
* of features, in the case where snip-edges == false; it also
* affects the return value of IsLastFrame().
*/
void InputFinished() const;
int32_t NumFramesReady() const;
/** Note: IsLastFrame() will only ever return true if you have called
* InputFinished() (and this frame is the last frame).
*/
bool IsLastFrame(int32_t frame) const;
/** Get n frames starting from the given frame index.
*
* @param frame_index The starting frame index
* @param n Number of frames to get.
* @return Return a 2-D tensor of shape (n, feature_dim).
* which is flattened into a 1-D vector (flattened in in row major)
*/
std::vector<float> GetFrames(int32_t frame_index, int32_t n) const;
void Reset();
int32_t FeatureDim() const;
// Return a reference to the number of processed frames so far
// before subsampling..
// Initially, it is 0. It is always less than NumFramesReady().
//
// The returned reference is valid as long as this object is alive.
int32_t &GetNumProcessedFrames(); // It's reset after calling Reset()
int32_t GetNumFramesSinceStart() const;
int32_t &GetCurrentSegment();
void SetResult(const OnlineTransducerDecoderResult &r);
OnlineTransducerDecoderResult &GetResult();
void SetKeywordResult(const TransducerKeywordResult &r);
TransducerKeywordResult &GetKeywordResult(bool remove_duplicates = false);
void SetCtcResult(const OnlineCtcDecoderResult &r);
OnlineCtcDecoderResult &GetCtcResult();
void SetParaformerResult(const OnlineParaformerDecoderResult &r);
OnlineParaformerDecoderResult &GetParaformerResult();
void SetStates(std::vector<Ort::Value> states);
std::vector<Ort::Value> &GetStates();
void SetNeMoDecoderStates(std::vector<Ort::Value> decoder_states);
std::vector<Ort::Value> &GetNeMoDecoderStates();
/**
* Get the context graph corresponding to this stream.
*
* @return Return the context graph for this stream.
*/
const ContextGraphPtr &GetContextGraph() const;
// for online ctc decoder
void SetFasterDecoder(std::unique_ptr<kaldi_decoder::FasterDecoder> decoder);
kaldi_decoder::FasterDecoder *GetFasterDecoder() const;
int32_t &GetFasterDecoderProcessedFrames();
// for streaming paraformer
std::vector<float> &GetParaformerFeatCache();
std::vector<float> &GetParaformerEncoderOutCache();
std::vector<float> &GetParaformerAlphaCache();
private:
class Impl;
std::unique_ptr<Impl> impl_;
};
} // namespace sherpa_onnx
#endif // SHERPA_ONNX_CSRC_ONLINE_STREAM_H_