Fabio
c36510c21e
Some checks failed
Linux Build / Linux cpu (push) Has been cancelled
Linux Build / Linux cpu-clang (push) Has been cancelled
Linux Build / Linux vulkan (push) Has been cancelled
mac-build / macOS CPU (push) Has been cancelled
Nix Build / Nix (cpu) (push) Has been cancelled
Nix Build / Nix (vulkan) (push) Has been cancelled
windows-build / Windows CPU (push) Has been cancelled
40 lines
1.1 KiB
C++
40 lines
1.1 KiB
C++
#pragma once
|
|
|
|
#include "engine/models/sherpa_kws/assets.h"
|
|
#include "engine/framework/core/execution_context.h"
|
|
|
|
#include <memory>
|
|
#include <vector>
|
|
|
|
namespace engine::models::sherpa_kws {
|
|
|
|
struct SherpaSubsampledChunk {
|
|
std::vector<float> values;
|
|
int64_t frames = 0;
|
|
int64_t channels = 0;
|
|
};
|
|
|
|
// Native implementation of the exported streaming Conv2dSubsampling
|
|
// boundary. Keeping this model-specific avoids changing the shared
|
|
// ZipEnhancer implementation while the complete Zipformer2 runtime is
|
|
// validated.
|
|
class SherpaEncoderRuntime {
|
|
public:
|
|
SherpaEncoderRuntime(
|
|
std::shared_ptr<const SherpaKwsAssets> assets,
|
|
core::ExecutionContext & execution_context);
|
|
~SherpaEncoderRuntime();
|
|
|
|
SherpaEncoderRuntime(const SherpaEncoderRuntime &) = delete;
|
|
SherpaEncoderRuntime & operator=(const SherpaEncoderRuntime &) = delete;
|
|
|
|
SherpaSubsampledChunk encode_subsampled_chunk(
|
|
const std::vector<float> & features) const;
|
|
void reset();
|
|
|
|
private:
|
|
struct Impl;
|
|
std::unique_ptr<Impl> impl_;
|
|
};
|
|
|
|
} // namespace engine::models::sherpa_kws
|