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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
|
/*
* Copyright (c) 2021 Arm Limited. All rights reserved.
* SPDX-License-Identifier: Apache-2.0
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "UseCaseHandler.hpp"
#include "InputFiles.hpp"
#include "AsrClassifier.hpp"
#include "Wav2LetterModel.hpp"
#include "hal.h"
#include "Wav2LetterMfcc.hpp"
#include "AudioUtils.hpp"
#include "UseCaseCommonUtils.hpp"
#include "AsrResult.hpp"
#include "Wav2LetterPreprocess.hpp"
#include "Wav2LetterPostprocess.hpp"
#include "OutputDecode.hpp"
namespace arm {
namespace app {
/**
* @brief Helper function to increment current audio clip index.
* @param[in,out] ctx Pointer to the application context object.
**/
static void IncrementAppCtxClipIdx(ApplicationContext& ctx);
/**
* @brief Helper function to set the audio clip index.
* @param[in,out] ctx Pointer to the application context object.
* @param[in] idx Value to be set.
* @return true if index is set, false otherwise.
**/
static bool SetAppCtxClipIdx(ApplicationContext& ctx, uint32_t idx);
/**
* @brief Presents inference results using the data presentation
* object.
* @param[in] platform Reference to the hal platform object.
* @param[in] results Vector of classification results to be displayed.
* @return true if successful, false otherwise.
**/
static bool PresentInferenceResult(
hal_platform& platform,
const std::vector<arm::app::asr::AsrResult>& results);
/* Audio inference classification handler. */
bool ClassifyAudioHandler(ApplicationContext& ctx, uint32_t clipIndex, bool runAll)
{
constexpr uint32_t dataPsnTxtInfStartX = 20;
constexpr uint32_t dataPsnTxtInfStartY = 40;
auto& platform = ctx.Get<hal_platform&>("platform");
platform.data_psn->clear(COLOR_BLACK);
auto& profiler = ctx.Get<Profiler&>("profiler");
/* If the request has a valid size, set the audio index. */
if (clipIndex < NUMBER_OF_FILES) {
if (!SetAppCtxClipIdx(ctx, clipIndex)) {
return false;
}
}
/* Get model reference. */
auto& model = ctx.Get<Model&>("model");
if (!model.IsInited()) {
printf_err("Model is not initialised! Terminating processing.\n");
return false;
}
/* Get score threshold to be applied for the classifier (post-inference). */
auto scoreThreshold = ctx.Get<float>("scoreThreshold");
/* Get tensors. Dimensions of the tensor should have been verified by
* the callee. */
TfLiteTensor* inputTensor = model.GetInputTensor(0);
TfLiteTensor* outputTensor = model.GetOutputTensor(0);
const uint32_t inputRows = inputTensor->dims->data[arm::app::Wav2LetterModel::ms_inputRowsIdx];
/* Populate MFCC related parameters. */
auto mfccParamsWinLen = ctx.Get<uint32_t>("frameLength");
auto mfccParamsWinStride = ctx.Get<uint32_t>("frameStride");
/* Populate ASR inference context and inner lengths for input. */
auto inputCtxLen = ctx.Get<uint32_t>("ctxLen");
const uint32_t inputInnerLen = inputRows - (2 * inputCtxLen);
/* Audio data stride corresponds to inputInnerLen feature vectors. */
const uint32_t audioParamsWinLen = (inputRows - 1) * mfccParamsWinStride + (mfccParamsWinLen);
const uint32_t audioParamsWinStride = inputInnerLen * mfccParamsWinStride;
const float audioParamsSecondsPerSample = (1.0/audio::Wav2LetterMFCC::ms_defaultSamplingFreq);
/* Get pre/post-processing objects. */
auto& prep = ctx.Get<audio::asr::Preprocess&>("preprocess");
auto& postp = ctx.Get<audio::asr::Postprocess&>("postprocess");
/* Set default reduction axis for post-processing. */
const uint32_t reductionAxis = arm::app::Wav2LetterModel::ms_outputRowsIdx;
/* Audio clip start index. */
auto startClipIdx = ctx.Get<uint32_t>("clipIndex");
/* Loop to process audio clips. */
do {
/* Get current audio clip index. */
auto currentIndex = ctx.Get<uint32_t>("clipIndex");
/* Get the current audio buffer and respective size. */
const int16_t* audioArr = get_audio_array(currentIndex);
const uint32_t audioArrSize = get_audio_array_size(currentIndex);
if (!audioArr) {
printf_err("Invalid audio array pointer\n");
return false;
}
/* Audio clip must have enough samples to produce 1 MFCC feature. */
if (audioArrSize < mfccParamsWinLen) {
printf_err("Not enough audio samples, minimum needed is %" PRIu32 "\n",
mfccParamsWinLen);
return false;
}
/* Initialise an audio slider. */
auto audioDataSlider = audio::FractionalSlidingWindow<const int16_t>(
audioArr,
audioArrSize,
audioParamsWinLen,
audioParamsWinStride);
/* Declare a container for results. */
std::vector<arm::app::asr::AsrResult> results;
/* Display message on the LCD - inference running. */
std::string str_inf{"Running inference... "};
platform.data_psn->present_data_text(
str_inf.c_str(), str_inf.size(),
dataPsnTxtInfStartX, dataPsnTxtInfStartY, 0);
info("Running inference on audio clip %" PRIu32 " => %s\n", currentIndex,
get_filename(currentIndex));
size_t inferenceWindowLen = audioParamsWinLen;
/* Start sliding through audio clip. */
while (audioDataSlider.HasNext()) {
/* If not enough audio see how much can be sent for processing. */
size_t nextStartIndex = audioDataSlider.NextWindowStartIndex();
if (nextStartIndex + audioParamsWinLen > audioArrSize) {
inferenceWindowLen = audioArrSize - nextStartIndex;
}
const int16_t* inferenceWindow = audioDataSlider.Next();
info("Inference %zu/%zu\n", audioDataSlider.Index() + 1,
static_cast<size_t>(ceilf(audioDataSlider.FractionalTotalStrides() + 1)));
/* Calculate MFCCs, deltas and populate the input tensor. */
prep.Invoke(inferenceWindow, inferenceWindowLen, inputTensor);
/* Run inference over this audio clip sliding window. */
if (!RunInference(model, profiler)) {
return false;
}
/* Post-process. */
postp.Invoke(outputTensor, reductionAxis, !audioDataSlider.HasNext());
/* Get results. */
std::vector<ClassificationResult> classificationResult;
auto& classifier = ctx.Get<AsrClassifier&>("classifier");
classifier.GetClassificationResults(
outputTensor, classificationResult,
ctx.Get<std::vector<std::string>&>("labels"), 1);
results.emplace_back(asr::AsrResult(classificationResult,
(audioDataSlider.Index() *
audioParamsSecondsPerSample *
audioParamsWinStride),
audioDataSlider.Index(), scoreThreshold));
#if VERIFY_TEST_OUTPUT
arm::app::DumpTensor(outputTensor,
outputTensor->dims->data[arm::app::Wav2LetterModel::ms_outputColsIdx]);
#endif /* VERIFY_TEST_OUTPUT */
}
/* Erase. */
str_inf = std::string(str_inf.size(), ' ');
platform.data_psn->present_data_text(
str_inf.c_str(), str_inf.size(),
dataPsnTxtInfStartX, dataPsnTxtInfStartY, 0);
ctx.Set<std::vector<arm::app::asr::AsrResult>>("results", results);
if (!PresentInferenceResult(platform, results)) {
return false;
}
profiler.PrintProfilingResult();
IncrementAppCtxClipIdx(ctx);
} while (runAll && ctx.Get<uint32_t>("clipIndex") != startClipIdx);
return true;
}
static void IncrementAppCtxClipIdx(ApplicationContext& ctx)
{
auto curAudioIdx = ctx.Get<uint32_t>("clipIndex");
if (curAudioIdx + 1 >= NUMBER_OF_FILES) {
ctx.Set<uint32_t>("clipIndex", 0);
return;
}
++curAudioIdx;
ctx.Set<uint32_t>("clipIndex", curAudioIdx);
}
static bool SetAppCtxClipIdx(ApplicationContext& ctx, uint32_t idx)
{
if (idx >= NUMBER_OF_FILES) {
printf_err("Invalid idx %" PRIu32 " (expected less than %u)\n",
idx, NUMBER_OF_FILES);
return false;
}
ctx.Set<uint32_t>("clipIndex", idx);
return true;
}
static bool PresentInferenceResult(hal_platform& platform,
const std::vector<arm::app::asr::AsrResult>& results)
{
constexpr uint32_t dataPsnTxtStartX1 = 20;
constexpr uint32_t dataPsnTxtStartY1 = 60;
constexpr bool allow_multiple_lines = true;
platform.data_psn->set_text_color(COLOR_GREEN);
info("Final results:\n");
info("Total number of inferences: %zu\n", results.size());
/* Results from multiple inferences should be combined before processing. */
std::vector<arm::app::ClassificationResult> combinedResults;
for (auto& result : results) {
combinedResults.insert(combinedResults.end(),
result.m_resultVec.begin(),
result.m_resultVec.end());
}
/* Get each inference result string using the decoder. */
for (const auto & result : results) {
std::string infResultStr = audio::asr::DecodeOutput(result.m_resultVec);
info("For timestamp: %f (inference #: %" PRIu32 "); label: %s\n",
result.m_timeStamp, result.m_inferenceNumber,
infResultStr.c_str());
}
/* Get the decoded result for the combined result. */
std::string finalResultStr = audio::asr::DecodeOutput(combinedResults);
platform.data_psn->present_data_text(
finalResultStr.c_str(), finalResultStr.size(),
dataPsnTxtStartX1, dataPsnTxtStartY1,
allow_multiple_lines);
info("Complete recognition: %s\n", finalResultStr.c_str());
return true;
}
} /* namespace app */
} /* namespace arm */
|
