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
111 lines
4.6 KiB
C++
111 lines
4.6 KiB
C++
#pragma once
|
|
|
|
#include "engine/framework/core/execution_context.h"
|
|
#include "engine/framework/runtime/cache_slots.h"
|
|
#include "engine/framework/runtime/session_base.h"
|
|
#include "engine/framework/runtime/streaming_audio.h"
|
|
#include "engine/framework/runtime/streaming_tts_session.h"
|
|
#include "engine/models/voxcpm2/assets.h"
|
|
#include "engine/models/voxcpm2/audiovae.h"
|
|
#include "engine/models/voxcpm2/generator.h"
|
|
|
|
#include <cstddef>
|
|
#include <memory>
|
|
#include <optional>
|
|
#include <string>
|
|
|
|
namespace engine::models::voxcpm2 {
|
|
|
|
class VoxCPM2SessionBase : public runtime::RuntimeSessionBase {
|
|
public:
|
|
VoxCPM2SessionBase(runtime::TaskSpec task, runtime::SessionOptions options,
|
|
std::shared_ptr<const VoxCPM2Assets> assets);
|
|
~VoxCPM2SessionBase() override;
|
|
|
|
protected:
|
|
std::string family_impl() const;
|
|
runtime::VoiceTaskKind task_kind_impl() const;
|
|
runtime::RunMode run_mode_impl() const;
|
|
void prepare_impl(const runtime::SessionPreparationRequest &request);
|
|
|
|
struct EncodedPromptCacheKey {
|
|
std::string prompt_text;
|
|
std::optional<runtime::AudioBuffer> prompt_audio;
|
|
std::optional<runtime::AudioBuffer> reference_audio;
|
|
};
|
|
|
|
struct EncodedPromptCacheKeyEqual {
|
|
bool operator()(const EncodedPromptCacheKey &lhs,
|
|
const EncodedPromptCacheKey &rhs) const;
|
|
};
|
|
|
|
struct EncodedPromptCacheEntry {
|
|
VoxCPM2EncodedPrompt encoded;
|
|
};
|
|
|
|
VoxCPM2GenerationOptions
|
|
generation_options_from_request(const runtime::TaskRequest &request) const;
|
|
void validate_request(const runtime::TaskRequest &request) const;
|
|
const VoxCPM2EncodedPrompt *encoded_prompt_for_request(
|
|
const std::optional<runtime::AudioBuffer> &prompt_audio,
|
|
const std::string &prompt_text,
|
|
const std::optional<runtime::AudioBuffer> &reference_audio);
|
|
/** request.audio_input, or with `ultimate_clone` the voice reference itself (with its transcript as prompt_text). */
|
|
const std::optional<runtime::AudioBuffer> &prompt_audio_for_request(
|
|
const runtime::TaskRequest &request,
|
|
const std::optional<runtime::AudioBuffer> &reference_audio,
|
|
const std::string &prompt_text) const;
|
|
|
|
runtime::TaskResult run_offline_request(const runtime::TaskRequest &request);
|
|
runtime::TaskResult run_streaming_request(
|
|
const runtime::TaskRequest &request,
|
|
const runtime::StreamEventCallback &stream_event_sink = nullptr);
|
|
void release_request_runtime_memory();
|
|
|
|
runtime::TaskSpec task_;
|
|
std::shared_ptr<const VoxCPM2Assets> assets_;
|
|
VoxCPM2FeatureGeneratorConfig generator_config_;
|
|
VoxCPM2AudioVAEDecoderConfig decoder_config_;
|
|
// Set by voxcpm2.exec_plan.audiovae.params.threads: the AudioVAE runs on its own CPU backend
|
|
// with its own threads, so streaming can decode a patch window while the
|
|
// generator works on the next patch. Declared before decoder_ so the decoder
|
|
// is destroyed first.
|
|
std::unique_ptr<engine::core::ExecutionContext> decoder_execution_context_;
|
|
std::unique_ptr<VoxCPM2FeatureGeneratorRuntime> generator_;
|
|
std::unique_ptr<VoxCPM2AudioVAEDecoderRuntime> decoder_;
|
|
runtime::CacheSlots<EncodedPromptCacheKey, EncodedPromptCacheEntry,
|
|
EncodedPromptCacheKeyEqual>
|
|
encoded_prompt_cache_;
|
|
std::optional<EncodedPromptCacheEntry> uncached_encoded_prompt_;
|
|
runtime::StreamingAudioController<float> streaming_audio_;
|
|
};
|
|
|
|
class VoxCPM2OfflineSession final : public VoxCPM2SessionBase,
|
|
public runtime::IOfflineVoiceTaskSession {
|
|
public:
|
|
VoxCPM2OfflineSession(runtime::TaskSpec task, runtime::SessionOptions options,
|
|
std::shared_ptr<const VoxCPM2Assets> assets);
|
|
|
|
std::string family() const override;
|
|
runtime::VoiceTaskKind task_kind() const override;
|
|
runtime::RunMode run_mode() const override;
|
|
void prepare(const runtime::SessionPreparationRequest &request) override;
|
|
runtime::TaskResult run(const runtime::TaskRequest &request) override;
|
|
};
|
|
|
|
class VoxCPM2StreamingSession final : public VoxCPM2SessionBase,
|
|
public runtime::StreamingTtsSessionBase {
|
|
public:
|
|
VoxCPM2StreamingSession(runtime::TaskSpec task, runtime::SessionOptions options,
|
|
std::shared_ptr<const VoxCPM2Assets> assets);
|
|
|
|
std::string family() const override;
|
|
runtime::VoiceTaskKind task_kind() const override;
|
|
runtime::RunMode run_mode() const override;
|
|
void prepare(const runtime::SessionPreparationRequest &request) override;
|
|
private:
|
|
runtime::TaskResult generate_stream(const runtime::TaskRequest &request) override;
|
|
void reset_stream_state() override { streaming_audio_.reset(); }
|
|
};
|
|
|
|
} // namespace engine::models::voxcpm2
|