George Gekov | 23c2627 | 2021-08-16 11:32:10 +0100 | [diff] [blame] | 1 | // |
| 2 | // Copyright © 2021 Arm Ltd and Contributors. All rights reserved. |
| 3 | // SPDX-License-Identifier: MIT |
| 4 | // |
| 5 | #ifndef KEYWORD_SPOTTING_EXAMPLE_DSCNNPREPROCESSOR_HPP |
| 6 | #define KEYWORD_SPOTTING_EXAMPLE_DSCNNPREPROCESSOR_HPP |
| 7 | |
| 8 | #include <numeric> |
| 9 | #include "DsCnnMfcc.hpp" |
| 10 | |
| 11 | namespace kws |
| 12 | { |
| 13 | class DsCNNPreprocessor |
| 14 | { |
| 15 | public: |
| 16 | DsCNNPreprocessor(uint32_t windowLen, uint32_t windowStride, |
| 17 | std::unique_ptr<DsCnnMFCC> mfccInst); |
| 18 | |
| 19 | /** |
| 20 | * @brief Calculates the features required from audio data. This |
| 21 | * includes MFCC, first and second order deltas, |
| 22 | * normalisation and finally, quantisation. The tensor is |
| 23 | * populated with feature from a given window placed along |
| 24 | * in a single row. |
| 25 | * @param[in] audioData pointer to the first element of audio data |
| 26 | * @param[in] output output to be populated |
| 27 | * @return true if successful, false in case of error. |
| 28 | */ |
| 29 | std::vector<int8_t> Invoke(const float* audioData, |
| 30 | size_t dataSize, |
| 31 | int quantOffset, |
| 32 | float quantScale) ; |
| 33 | |
| 34 | uint32_t m_windowLen; // Window length for MFCC |
| 35 | uint32_t m_windowStride; // Window stride len for MFCC |
| 36 | std::unique_ptr<MFCC> m_mfcc; |
| 37 | }; |
| 38 | } // namespace kws |
| 39 | #endif //KEYWORD_SPOTTING_EXAMPLE_DSCNNPREPROCESSOR_HPP |