40 lines
1.0 KiB
C++
40 lines
1.0 KiB
C++
// sherpa-onnx/csrc/vad-model.h
|
|
//
|
|
// Copyright (c) 2023 Xiaomi Corporation
|
|
#ifndef SHERPA_ONNX_CSRC_VAD_MODEL_H_
|
|
#define SHERPA_ONNX_CSRC_VAD_MODEL_H_
|
|
|
|
#include <memory>
|
|
|
|
#include "sherpa-onnx/csrc/vad-model-config.h"
|
|
|
|
namespace sherpa_onnx {
|
|
|
|
class VadModel {
|
|
public:
|
|
virtual ~VadModel() = default;
|
|
|
|
static std::unique_ptr<VadModel> Create(const VadModelConfig &config);
|
|
|
|
// reset the internal model states
|
|
virtual void Reset() = 0;
|
|
|
|
/**
|
|
* @param samples Pointer to a 1-d array containing audio samples.
|
|
* Each sample should be normalized to the range [-1, 1].
|
|
* @param n Number of samples. Should be equal to WindowSize()
|
|
*
|
|
* @return Return true if speech is detected. Return false otherwise.
|
|
*/
|
|
virtual bool IsSpeech(const float *samples, int32_t n) = 0;
|
|
|
|
virtual int32_t WindowSize() const = 0;
|
|
|
|
virtual int32_t MinSilenceDurationSamples() const = 0;
|
|
virtual int32_t MinSpeechDurationSamples() const = 0;
|
|
};
|
|
|
|
} // namespace sherpa_onnx
|
|
|
|
#endif // SHERPA_ONNX_CSRC_VAD_MODEL_H_
|