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
81 lines
2.8 KiB
C++
81 lines
2.8 KiB
C++
#pragma once
|
|
|
|
#include "engine/framework/assets/resource_bundle.h"
|
|
#include "engine/framework/assets/tensor_source.h"
|
|
#include "engine/framework/tokenizers/sentencepiece.h"
|
|
|
|
#include <cstdint>
|
|
#include <filesystem>
|
|
#include <memory>
|
|
#include <optional>
|
|
#include <string>
|
|
#include <vector>
|
|
|
|
namespace engine::models::pocket_tts {
|
|
|
|
struct PocketTTSModelConfig {
|
|
float default_temperature = 0.7F;
|
|
int sample_rate = 24000;
|
|
float frame_rate = 12.5F;
|
|
float mimi_frame_rate = 12.5F;
|
|
int64_t flow_layers = 6;
|
|
int64_t flow_depth = 6;
|
|
int64_t flow_dim = 1024;
|
|
int64_t flow_heads = 16;
|
|
int64_t flow_hidden_size = 512;
|
|
int64_t flow_intermediate_size = 4096;
|
|
int64_t latent_dim = 32;
|
|
int64_t mimi_layers = 2;
|
|
int64_t mimi_dim = 512;
|
|
int64_t mimi_heads = 8;
|
|
int64_t mimi_intermediate_size = 2048;
|
|
int64_t mimi_inner_dim = 32;
|
|
int64_t mimi_outer_dim = 512;
|
|
int64_t mimi_context = 250;
|
|
int64_t mimi_seanet_dimension = 512;
|
|
int64_t mimi_base_filters = 64;
|
|
int64_t mimi_encoder_upsample_stride = 16;
|
|
bool pad_with_spaces_for_short_inputs = false;
|
|
bool remove_semicolons = false;
|
|
bool insert_bos_before_voice = false;
|
|
std::optional<int> model_recommended_frames_after_eos = std::nullopt;
|
|
// Added to the word-count guess (3 or 1) for the default frames_after_eos, which counts
|
|
// the EOS frame itself.
|
|
int frames_after_eos_offset = 2;
|
|
float default_eos_threshold = -4.0F;
|
|
bool time_embedding_plain_rms = false;
|
|
};
|
|
|
|
struct PocketTTSHostWeights {
|
|
assets::TensorDataF32 conditioner_embedding_table;
|
|
std::vector<float> bos_emb;
|
|
std::optional<assets::TensorDataF32> bos_before_voice;
|
|
std::vector<float> emb_mean;
|
|
std::vector<float> emb_std;
|
|
// Constant added to input_linear on every generated frame, for checkpoints whose summed
|
|
// conditioners are folded at load time.
|
|
std::optional<std::vector<float>> input_linear_bias;
|
|
// Projects `speaker_wavs` voice latents [frames, channels] to flow_dim on the host.
|
|
std::optional<assets::TensorDataF32> speaker_proj;
|
|
};
|
|
|
|
struct PocketTTSAssets {
|
|
assets::ResourceBundle resources;
|
|
std::shared_ptr<const assets::TensorSource> model_weights;
|
|
std::filesystem::path voice_asset_root;
|
|
PocketTTSHostWeights host_weights;
|
|
PocketTTSModelConfig model_config;
|
|
std::vector<tokenizers::SentencePiecePiece> tokenizer_pieces;
|
|
std::string language;
|
|
std::string family = "pocket_tts";
|
|
// Voice used when a request selects none.
|
|
std::optional<std::filesystem::path> default_voice_path;
|
|
};
|
|
|
|
std::shared_ptr<const PocketTTSAssets> load_pocket_tts_assets(
|
|
const std::filesystem::path & model_path,
|
|
std::string language);
|
|
|
|
std::shared_ptr<const PocketTTSAssets> load_gradium_phonon_assets(const std::filesystem::path & model_path);
|
|
|
|
} // namespace engine::models::pocket_tts
|