Merge pull request #189 from tuna2134/renovate/ureq-3.x-lockfile

Fix example

.env.sample

@@ -3,4 +3,5 @@ MODEL_PATH=models/tsukuyomi.sbv2
 MODELS_PATH=models
 TOKENIZER_PATH=models/tokenizer.json
 ADDR=localhost:3000
-RUST_LOG=warn
+RUST_LOG=warn
+HOLDER_MAX_LOADED_MODElS=20

.github/workflows/CI.Dockerfile

@@ -0,0 +1,4 @@
+FROM ubuntu:latest
+RUN apt update && apt install openssl libssl-dev curl pkg-config software-properties-common -y && add-apt-repository ppa:deadsnakes/ppa && apt update && apt install python3.7 python3.8 python3.9 python3.10 python3.11 python3.12 python3.13 python3-pip python3 -y
+ENV PIP_BREAK_SYSTEM_PACKAGES=1
+RUN mkdir -p /root/.cache/sbv2 && curl https://huggingface.co/neody/sbv2-api-assets/resolve/main/dic/all.bin -o /root/.cache/sbv2/all.bin -L

.github/workflows/CI.yml

@@ -1,39 +1,35 @@
-# This file is autogenerated by maturin v1.7.1
-# To update, run
-#
-#    maturin generate-ci github
-#
 name: CI
 
 on:
   push:
     branches:
       - main
-      - master
     tags:
       - '*'
   pull_request:
   workflow_dispatch:
 
 permissions:
-  contents: read
+  contents: write
   id-token: write
+  packages: write
 
 jobs:
-  linux:
+  python-linux:
     runs-on: ${{ matrix.platform.runner }}
     strategy:
       matrix:
         platform:
           - runner: ubuntu-latest
             target: x86_64
-          - runner: ubuntu-latest
+          - runner: ubuntu-24.04-arm
             target: aarch64
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5
         with:
           python-version: 3.x
+      - run: docker build . -f .github/workflows/CI.Dockerfile --tag ci
       - name: Build wheels
         uses: PyO3/maturin-action@v1
         with:
@@ -41,14 +37,15 @@ jobs:
           args: --release --out dist --find-interpreter
           sccache: 'true'
           manylinux: auto
-          working-directory: sbv2_bindings
+          container: ci
+          working-directory: ./crates/sbv2_bindings
       - name: Upload wheels
         uses: actions/upload-artifact@v4
         with:
           name: wheels-linux-${{ matrix.platform.target }}
-          path: sbv2_bindings/dist
+          path: ./crates/sbv2_bindings/dist
 
-  windows:
+  python-windows:
     runs-on: ${{ matrix.platform.runner }}
     strategy:
       matrix:
@@ -67,20 +64,18 @@ jobs:
           target: ${{ matrix.platform.target }}
           args: --release --out dist --find-interpreter
           sccache: 'true'
-          working-directory: sbv2_bindings
+          working-directory: ./crates/sbv2_bindings
       - name: Upload wheels
         uses: actions/upload-artifact@v4
         with:
           name: wheels-windows-${{ matrix.platform.target }}
-          path: sbv2_bindings/dist
+          path: ./crates/sbv2_bindings/dist
 
-  macos:
+  python-macos:
     runs-on: ${{ matrix.platform.runner }}
     strategy:
       matrix:
         platform:
-          - runner: macos-12
-            target: x86_64
           - runner: macos-14
             target: aarch64
     steps:
@@ -94,14 +89,14 @@ jobs:
           target: ${{ matrix.platform.target }}
           args: --release --out dist --find-interpreter
           sccache: 'true'
-          working-directory: sbv2_bindings
+          working-directory: ./crates/sbv2_bindings
       - name: Upload wheels
         uses: actions/upload-artifact@v4
         with:
           name: wheels-macos-${{ matrix.platform.target }}
-          path: sbv2_bindings/dist
+          path: ./crates/sbv2_bindings/dist
 
-  sdist:
+  python-sdist:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
@@ -110,54 +105,116 @@ jobs:
         with:
           command: sdist
           args: --out dist
-          working-directory: sbv2_bindings
+          working-directory: ./crates/sbv2_bindings
       - name: Upload sdist
         uses: actions/upload-artifact@v4
         with:
           name: wheels-sdist
-          path: sbv2_bindings/dist
+          path: ./crates/sbv2_bindings/dist
+  python-wheel:
+    name: Wheel Upload
+    runs-on: ubuntu-latest
+    needs: [python-linux, python-windows, python-macos, python-sdist]
+    env:
+      GH_TOKEN: ${{ github.token }}
+    steps:
+      - uses: actions/checkout@v4
+      - run: gh run download ${{ github.run_id }} -p wheels-*
+      - name: release
+        run: |
+          gh release create commit-${GITHUB_SHA:0:8} --prerelease wheels-*/*
 
-  release:
+  python-release:
     name: Release
     runs-on: ubuntu-latest
     if: "startsWith(github.ref, 'refs/tags/')"
-    needs: [linux, windows, macos, sdist]
+    needs: [python-linux, python-windows, python-macos, python-sdist]
     environment: release
+    env:
+      GH_TOKEN: ${{ github.token }}
     steps:
-      - uses: actions/download-artifact@v4
+      - uses: actions/checkout@v4
+      - run: gh run download ${{ github.run_id }} -p wheels-*
       - name: Publish to PyPI
         uses: PyO3/maturin-action@v1
         with:
           command: upload
           args: --non-interactive --skip-existing wheels-*/*
 
-  push-docker:
-    runs-on: ubuntu-latest
-    if: "startsWith(github.ref, 'refs/tags/')"
-    permissions:
-      contents: read
-      packages: write
+  docker:
+    runs-on: ${{ matrix.machine.runner }}
     strategy:
+      fail-fast: false
       matrix:
+        machine:
+          - platform: amd64
+            runner: ubuntu-latest
+          - platform: arm64
+            runner: ubuntu-24.04-arm
         tag: [cpu, cuda]
     steps:
-      - uses: actions/checkout@v4
-      - name: Set up QEMU
-        uses: docker/setup-qemu-action@v3
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      - name: Login to GitHub Container Registry
+      - name: Prepare
+        run: |
+          platform=${{ matrix.machine.platform }}
+          echo "PLATFORM_PAIR=${platform//\//-}" >> $GITHUB_ENV
+
+      - name: Docker meta
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: |
+            ghcr.io/${{ github.repository }}
+
+      - name: Login to GHCR
         uses: docker/login-action@v3
         with:
           registry: ghcr.io
-          username: ${{ github.actor }}
+          username: ${{ github.repository_owner }}
           password: ${{ secrets.GITHUB_TOKEN }}
-      - name: Build and push image
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Build and push by digest
+        id: build
         uses: docker/build-push-action@v6
         with:
-          context: .
+          labels: ${{ steps.meta.outputs.labels }}
+          file: ./scripts/docker/${{ matrix.tag }}.Dockerfile
           push: true
           tags: |
-            ghcr.io/${{ github.repository }}:${{ matrix.tag }}
-          file: docker/${{ matrix.tag }}.Dockerfile
-          platforms: linux/amd64, linux/arm64
+            ghcr.io/${{ github.repository }}:latest-${{ matrix.tag }}-${{ matrix.machine.platform }}
+
+  docker-merge:
+    runs-on: ubuntu-latest
+    needs:
+      - docker
+    steps:
+      - name: Download digests
+        uses: actions/download-artifact@v4
+        with:
+          path: ${{ runner.temp }}/digests
+          pattern: digests-*
+          merge-multiple: true
+
+      - name: Login to GHCR
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.repository_owner }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Merge
+        run: |
+          docker buildx imagetools create -t ghcr.io/${{ github.repository }}:cuda \
+              ghcr.io/${{ github.repository }}:latest-cuda-amd64 \
+              ghcr.io/${{ github.repository }}:latest-cuda-arm64
+          docker buildx imagetools create -t ghcr.io/${{ github.repository }}:cpu \
+              ghcr.io/${{ github.repository }}:latest-cpu-amd64 \
+              ghcr.io/${{ github.repository }}:latest-cpu-arm64

.gitignore

@@ -1,8 +1,10 @@
-target
+target/
 models/
 !models/.gitkeep
 venv/
 .env
-output.wav
-node_modules
-dist/
+*.wav
+node_modules/
+dist/
+*.csv
+*.bin

Cargo.toml

@@ -1,15 +1,25 @@
 [workspace]
 resolver = "2"
-members = ["sbv2_api", "sbv2_core", "sbv2_bindings", "sbv2_wasm"]
+members = ["./crates/sbv2_api", "./crates/sbv2_core", "./crates/sbv2_bindings", "./crates/sbv2_wasm"]
+
+[workspace.package]
+version = "0.2.0-alpha6"
+edition = "2021"
+description = "Style-Bert-VITSの推論ライブラリ"
+license = "MIT"
+readme = "./README.md"
+repository = "https://github.com/tuna2134/sbv2-api"
+documentation = "https://docs.rs/sbv2_core"
 
 [workspace.dependencies]
-anyhow = "1.0.86"
+anyhow = "1.0.96"
 dotenvy = "0.15.7"
-env_logger = "0.11.5"
+env_logger = "0.11.6"
 ndarray = "0.16.1"
-once_cell = "1.19.0"
+once_cell = "1.20.3"
 
 [profile.release]
-lto = true
-debug = false
 strip = true
+opt-level = "z"
+lto = true
+codegen-units = 1

README.md

@@ -3,6 +3,9 @@
 ## 注意：本バージョンはアルファ版です。
 安定版を利用したい場合は[こちら](https://github.com/tuna2134/sbv2-api/tree/v0.1.x)をご覧ください。
 
+## 注意: オプションの辞書はLGPLです。
+オプションの辞書を使用する場合、バイナリの内部の辞書部分について、LGPLが適用されます。
+
 ## プログラミングに詳しくない方向け
 
 [こちら](https://github.com/tuna2134/sbv2-gui?tab=readme-ov-file)を参照してください。
@@ -17,7 +20,7 @@ JP-Extra しか対応していません。(基本的に対応する予定もあ
 
 ## 変換方法
 
-[こちら](https://github.com/tuna2134/sbv2-api/tree/main/convert)を参照してください。
+[こちら](https://github.com/tuna2134/sbv2-api/tree/main/scripts/convert)を参照してください。
 
 ## Todo
 
@@ -29,23 +32,27 @@ JP-Extra しか対応していません。(基本的に対応する予定もあ
 - [x] GPU 対応(CUDA)
 - [x] GPU 対応(DirectML)
 - [x] GPU 対応(CoreML)
-- [ ] WASM 変換(依存ライブラリの関係により現在は不可)
+- [x] WASM 変換
 - [x] arm64のdockerサポート
+- [x] aivis形式のサポート
 - [ ] MeCabを利用する
 
 ## 構造説明
 
-- `sbv2_api` - 推論用 REST API
-- `sbv2_core` - 推論コア部分
-- `docker` - docker ビルドスクリプト
-- `convert` - onnx, sbv2フォーマットへの変換スクリプト
+- `crates/sbv2_api` - 推論用 REST API
+- `crates/sbv2_core` - 推論コア部分
+- `scripts/docker` - docker ビルドスクリプト
+- `scripts/convert` - onnx, sbv2フォーマットへの変換スクリプト
 
 ## プログラミングある程度できる人向けREST API起動方法
 
 ### models をインストール
 
-https://huggingface.co/googlefan/sbv2_onnx_models/tree/main
-の`tokenizer.json`,`debert.onnx`,`tsukuyomi.sbv2`を models フォルダに配置
+https://huggingface.co/neody/sbv2-api-assets/tree/main/deberta
+から`tokenizer.json`,`debert.onnx`
+https://huggingface.co/neody/sbv2-api-assets/tree/main/model
+から`tsukuyomi.sbv2`
+を models フォルダに配置
 
 ### .env ファイルの作成
 
@@ -114,8 +121,10 @@ curl http://localhost:3000/models
 - `ADDR` `localhost:3000`などのようにサーバー起動アドレスをコントロールできます。
 - `MODELS_PATH` sbv2モデルの存在するフォルダを指定できます。
 - `RUST_LOG` おなじみlog levelです。
+- `HOLDER_MAX_LOADED_MODElS` RAMにロードされるモデルの最大数を指定します。
 
 ## 謝辞
 
-- [litagin02/Style-Bert-VITS2](https://github.com/litagin02/Style-Bert-VITS2) - このコードの書くにあたり、ベースとなる部分を参考にさせていただきました。
+- [litagin02/Style-Bert-VITS2](https://github.com/litagin02/Style-Bert-VITS2) - このコードを書くにあたり、ベースとなる部分を参考にさせていただきました。
 - [Googlefan](https://github.com/Googlefan256) - 彼にモデルを ONNX ヘ変換および効率化をする方法を教わりました。
+- [Aivis Project](https://github.com/Aivis-Project/AivisSpeech-Engine) - 辞書部分

content.txt

@@ -1,14 +0,0 @@
-悪徳貴族として名高いヴェレット家の長男――オウガ・ヴェレットは転生者である。
-ブラック企業に勤め、過労死した彼には一つの夢があった。
-
-「可愛いハーレム作って、美味い物を食べる。領民の税金で楽して好き放題な生活を送ってみせる！」
-
-素晴らしき異世界ライフを夢見た彼は実現へ向けて、努力を始めた。
-ハーレムを築くためにいじめられてる平民の子を助けて恩を売ってやったり。
-労働力を手に入れるために多くの孤児を雇って教育したり。
-反乱を起きても鎮圧できるように魔法学院へ通って魔法を極める。
-
-「クックック……！　順調、順調！　未来は明るいなぁ！」
-
-――オウガはまだ知らない。
-楽な生活を送るためにしてきたことが評価され、世間から『聖者』様として呼ばれる未来を。

crates/sbv2_api/Cargo.toml

@@ -1,19 +1,24 @@
 [package]
 name = "sbv2_api"
-version = "0.2.0-alpha"
-edition = "2021"
+version.workspace = true
+edition.workspace = true
+description.workspace = true
+readme.workspace = true
+repository.workspace = true
+documentation.workspace = true
+license.workspace = true
 
 [dependencies]
 anyhow.workspace = true
-axum = "0.7.5"
+axum = "0.8.0"
 dotenvy.workspace = true
 env_logger.workspace = true
 log = "0.4.22"
-sbv2_core = { version = "0.2.0-alpha2", path = "../sbv2_core" }
+sbv2_core = { version = "0.2.0-alpha6", path = "../sbv2_core", features = ["aivmx"] }
 serde = { version = "1.0.210", features = ["derive"] }
 tokio = { version = "1.40.0", features = ["full"] }
 utoipa = { version = "5.0.0", features = ["axum_extras"] }
-utoipa-scalar = { version = "0.2.0", features = ["axum"] }
+utoipa-scalar = { version = "0.3.0", features = ["axum"] }
 
 [features]
 coreml = ["sbv2_core/coreml"]

crates/sbv2_api/src/main.rs

@@ -140,6 +140,20 @@ impl AppState {
                     log::warn!("Error loading {entry}: {e}");
                 };
                 log::info!("Loaded: {entry}");
+            } else if name.ends_with(".aivmx") {
+                let entry = &name[..name.len() - 6];
+                log::info!("Try loading: {entry}");
+                let aivmx_bytes = match fs::read(format!("{models}/{entry}.aivmx")).await {
+                    Ok(b) => b,
+                    Err(e) => {
+                        log::warn!("Error loading aivmx bytes from file {entry}: {e}");
+                        continue;
+                    }
+                };
+                if let Err(e) = tts_model.load_aivmx(entry, aivmx_bytes) {
+                    log::error!("Error loading {entry}: {e}");
+                }
+                log::info!("Loaded: {entry}");
             }
         }
         for entry in entries {

crates/sbv2_bindings/Cargo.toml

@@ -0,0 +1,24 @@
+[package]
+name = "sbv2_bindings"
+version.workspace = true
+edition.workspace = true
+description.workspace = true
+readme.workspace = true
+repository.workspace = true
+documentation.workspace = true
+license.workspace = true
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+[lib]
+name = "sbv2_bindings"
+crate-type = ["cdylib"]
+
+[dependencies]
+anyhow.workspace = true
+ndarray.workspace = true
+pyo3 = { version = "0.23.0", features = ["anyhow"] }
+sbv2_core = { path = "../sbv2_core", features = ["std"], default-features = false }
+
+[features]
+agpl_dict = ["sbv2_core/agpl_dict"]
+default = ["agpl_dict"]

crates/sbv2_bindings/src/sbv2.rs

@@ -107,7 +107,7 @@ impl TTSModel {
     /// style_vector : StyleVector
     ///     スタイルベクトル
     fn get_style_vector(
-        &self,
+        &mut self,
         ident: String,
         style_id: i32,
         weight: f32,
@@ -145,7 +145,7 @@ impl TTSModel {
         speaker_id: i64,
         sdp_ratio: f32,
         length_scale: f32,
-    ) -> anyhow::Result<Bound<PyBytes>> {
+    ) -> anyhow::Result<Bound<'p, PyBytes>> {
         let data = self.model.easy_synthesize(
             ident.as_str(),
             &text,
@@ -157,7 +157,7 @@ impl TTSModel {
                 ..Default::default()
             },
         )?;
-        Ok(PyBytes::new_bound(py, &data))
+        Ok(PyBytes::new(py, &data))
     }
 
     fn unload(&mut self, ident: String) -> bool {

crates/sbv2_core/Cargo.toml

@@ -1,38 +1,47 @@
 [package]
 name = "sbv2_core"
-description = "Style-Bert-VITSの推論ライブラリ"
-version = "0.2.0-alpha2"
-edition = "2021"
-license = "MIT"
-readme = "../README.md"
-repository = "https://github.com/tuna2134/sbv2-api"
-documentation = "https://docs.rs/sbv2_core"
+version.workspace = true
+edition.workspace = true
+description.workspace = true
+readme.workspace = true
+repository.workspace = true
+documentation.workspace = true
+license.workspace = true
 
 [dependencies]
 anyhow.workspace = true
+base64 = { version = "0.22.1", optional = true }
 dotenvy.workspace = true
 env_logger.workspace = true
 hound = "3.5.1"
-jpreprocess = { version = "0.10.0", features = ["naist-jdic"] }
+jpreprocess = { version = "0.12.0", features = ["naist-jdic"] }
 ndarray.workspace = true
+npyz = { version = "0.8.3", optional = true }
 num_cpus = "1.16.0"
 once_cell.workspace = true
-ort = { git = "https://github.com/pykeio/ort.git", version = "2.0.0-rc.8", optional = true }
+ort = { git = "https://github.com/pykeio/ort.git", version = "2.0.0-rc.9", optional = true }
 regex = "1.10.6"
 serde = { version = "1.0.210", features = ["derive"] }
 serde_json = "1.0.128"
 tar = "0.4.41"
-thiserror = "1.0.63"
-tokenizers = { version = "0.20.0", default-features = false }
+thiserror = "2.0.11"
+tokenizers = { version = "0.21.0", default-features = false }
 zstd = "0.13.2"
 
 [features]
 cuda = ["ort/cuda", "std"]
 cuda_tf32 = ["std", "cuda"]
+agpl_dict = []
 std = ["dep:ort", "tokenizers/progressbar", "tokenizers/onig", "tokenizers/esaxx_fast"]
 dynamic = ["ort/load-dynamic", "std"]
 directml = ["ort/directml", "std"]
 tensorrt = ["ort/tensorrt", "std"]
 coreml = ["ort/coreml", "std"]
-default = ["std"]
-no_std = ["tokenizers/unstable_wasm"]
+default = ["std", "agpl_dict"]
+no_std = ["tokenizers/unstable_wasm"]
+aivmx = ["npyz", "base64"]
+base64 = ["dep:base64"]
+
+[build-dependencies]
+dirs = "6.0.0"
+ureq = "3.0.6"

crates/sbv2_core/build.rs

@@ -0,0 +1,25 @@
+use dirs::home_dir;
+use std::env;
+use std::fs;
+use std::io::copy;
+use std::path::PathBuf;
+
+fn main() -> Result<(), Box<dyn std::error::Error>> {
+    let static_path = home_dir().unwrap().join(".cache/sbv2/all.bin");
+    let out_path = PathBuf::from(&env::var("OUT_DIR").unwrap()).join("all.bin");
+    println!("cargo:rerun-if-changed=build.rs");
+    if static_path.exists() {
+        if fs::hard_link(&static_path, &out_path).is_err() {
+            fs::copy(static_path, out_path).unwrap();
+        };
+    } else {
+        println!("cargo:warning=Downloading dictionary file...");
+        let mut response =
+            ureq::get("https://huggingface.co/neody/sbv2-api-assets/resolve/main/dic/all.bin")
+                .call()?;
+        let mut response = response.body_mut().as_reader();
+        let mut file = fs::File::create(&out_path)?;
+        copy(&mut response, &mut file)?;
+    }
+    Ok(())
+}

crates/sbv2_core/src/bert.rs

@@ -1,17 +1,18 @@
 use crate::error::Result;
 use ndarray::{Array2, Ix2};
-use ort::Session;
+use ort::session::Session;
+use ort::value::TensorRef;
 
 pub fn predict(
-    session: &Session,
+    session: &mut Session,
     token_ids: Vec<i64>,
     attention_masks: Vec<i64>,
 ) -> Result<Array2<f32>> {
     let outputs = session.run(
         ort::inputs! {
-            "input_ids" => Array2::from_shape_vec((1, token_ids.len()), token_ids).unwrap(),
-            "attention_mask" => Array2::from_shape_vec((1, attention_masks.len()), attention_masks).unwrap(),
-        }?
+            "input_ids" => TensorRef::from_array_view((vec![1, token_ids.len() as i64], token_ids.as_slice()))?,
+            "attention_mask" => TensorRef::from_array_view((vec![1, attention_masks.len() as i64], attention_masks.as_slice()))?,
+        }
     )?;
 
     let output = outputs["output"]

crates/sbv2_core/src/error.rs

@@ -6,6 +6,8 @@ pub enum Error {
     TokenizerError(#[from] tokenizers::Error),
     #[error("JPreprocess error: {0}")]
     JPreprocessError(#[from] jpreprocess::error::JPreprocessError),
+    #[error("Lindera error: {0}")]
+    LinderaError(String),
     #[cfg(feature = "std")]
     #[error("ONNX error: {0}")]
     OrtError(#[from] ort::Error),
@@ -21,6 +23,9 @@ pub enum Error {
     HoundError(#[from] hound::Error),
     #[error("model not found error")]
     ModelNotFoundError(String),
+    #[cfg(feature = "base64")]
+    #[error("base64 error")]
+    Base64Error(#[from] base64::DecodeError),
     #[error("other")]
     OtherError(String),
 }

crates/sbv2_core/src/jtalk.rs

@@ -1,21 +1,32 @@
 use crate::error::{Error, Result};
 use crate::mora::{MORA_KATA_TO_MORA_PHONEMES, VOWELS};
 use crate::norm::{replace_punctuation, PUNCTUATIONS};
-use jpreprocess::*;
+use jpreprocess::{kind, DefaultTokenizer, JPreprocess, SystemDictionaryConfig, UserDictionary};
 use once_cell::sync::Lazy;
 use regex::Regex;
 use std::cmp::Reverse;
 use std::collections::HashSet;
 use std::sync::Arc;
 
-type JPreprocessType = JPreprocess<DefaultFetcher>;
+type JPreprocessType = JPreprocess<DefaultTokenizer>;
+
+#[cfg(feature = "agpl_dict")]
+fn agpl_dict() -> Result<Option<UserDictionary>> {
+    Ok(Some(
+        UserDictionary::load(include_bytes!(concat!(env!("OUT_DIR"), "/all.bin")))
+            .map_err(|e| Error::LinderaError(e.to_string()))?,
+    ))
+}
+
+#[cfg(not(feature = "agpl_dict"))]
+fn agpl_dict() -> Result<Option<UserDictionary>> {
+    Ok(None)
+}
 
 fn initialize_jtalk() -> Result<JPreprocessType> {
-    let config = JPreprocessConfig {
-        dictionary: SystemDictionaryConfig::Bundled(kind::JPreprocessDictionaryKind::NaistJdic),
-        user_dictionary: None,
-    };
-    let jpreprocess = JPreprocess::from_config(config)?;
+    let sdic =
+        SystemDictionaryConfig::Bundled(kind::JPreprocessDictionaryKind::NaistJdic).load()?;
+    let jpreprocess = JPreprocess::with_dictionaries(sdic, agpl_dict()?);
     Ok(jpreprocess)
 }
 

crates/sbv2_core/src/main.rs

@@ -6,7 +6,7 @@ fn main_inner() -> anyhow::Result<()> {
     use sbv2_core::tts;
     dotenvy::dotenv_override().ok();
     env_logger::init();
-    let text = fs::read_to_string("content.txt")?;
+    let text = "今日の天気は快晴です。";
     let ident = "aaa";
     let mut tts_holder = tts::TTSModelHolder::new(
         &fs::read(env::var("BERT_MODEL_PATH")?)?,
@@ -15,7 +15,20 @@ fn main_inner() -> anyhow::Result<()> {
             .ok()
             .and_then(|x| x.parse().ok()),
     )?;
-    tts_holder.load_sbv2file(ident, fs::read(env::var("MODEL_PATH")?)?)?;
+    let mp = env::var("MODEL_PATH")?;
+    let b = fs::read(&mp)?;
+    #[cfg(not(feature = "aivmx"))]
+    {
+        tts_holder.load_sbv2file(ident, b)?;
+    }
+    #[cfg(feature = "aivmx")]
+    {
+        if mp.ends_with(".sbv2") {
+            tts_holder.load_sbv2file(ident, b)?;
+        } else {
+            tts_holder.load_aivmx(ident, b)?;
+        }
+    }
 
     let audio =
         tts_holder.easy_synthesize(ident, &text, 0, 0, tts::SynthesizeOptions::default())?;
@@ -23,6 +36,7 @@ fn main_inner() -> anyhow::Result<()> {
 
     Ok(())
 }
+
 #[cfg(not(feature = "std"))]
 fn main_inner() -> anyhow::Result<()> {
     Ok(())

crates/sbv2_core/src/model.rs

@@ -0,0 +1,111 @@
+use crate::error::Result;
+use ndarray::{array, Array1, Array2, Array3, Axis, Ix3};
+use ort::session::{builder::GraphOptimizationLevel, Session};
+
+#[allow(clippy::vec_init_then_push, unused_variables)]
+pub fn load_model<P: AsRef<[u8]>>(model_file: P, bert: bool) -> Result<Session> {
+    let mut exp = Vec::new();
+    #[cfg(feature = "tensorrt")]
+    {
+        if bert {
+            exp.push(
+                ort::execution_providers::TensorRTExecutionProvider::default()
+                    .with_fp16(true)
+                    .with_profile_min_shapes("input_ids:1x1,attention_mask:1x1")
+                    .with_profile_max_shapes("input_ids:1x100,attention_mask:1x100")
+                    .with_profile_opt_shapes("input_ids:1x25,attention_mask:1x25")
+                    .build(),
+            );
+        }
+    }
+    #[cfg(feature = "cuda")]
+    {
+        #[allow(unused_mut)]
+        let mut cuda = ort::execution_providers::CUDAExecutionProvider::default()
+            .with_conv_algorithm_search(
+                ort::execution_providers::cuda::CUDAExecutionProviderCuDNNConvAlgoSearch::Default,
+            );
+        #[cfg(feature = "cuda_tf32")]
+        {
+            cuda = cuda.with_tf32(true);
+        }
+        exp.push(cuda.build());
+    }
+    #[cfg(feature = "directml")]
+    {
+        exp.push(ort::execution_providers::DirectMLExecutionProvider::default().build());
+    }
+    #[cfg(feature = "coreml")]
+    {
+        exp.push(ort::execution_providers::CoreMLExecutionProvider::default().build());
+    }
+    exp.push(ort::execution_providers::CPUExecutionProvider::default().build());
+    Ok(Session::builder()?
+        .with_execution_providers(exp)?
+        .with_optimization_level(GraphOptimizationLevel::Level3)?
+        .with_intra_threads(num_cpus::get_physical())?
+        .with_parallel_execution(true)?
+        .with_inter_threads(num_cpus::get_physical())?
+        .commit_from_memory(model_file.as_ref())?)
+}
+
+#[allow(clippy::too_many_arguments)]
+pub fn synthesize(
+    session: &mut Session,
+    bert_ori: Array2<f32>,
+    x_tst: Array1<i64>,
+    mut spk_ids: Array1<i64>,
+    tones: Array1<i64>,
+    lang_ids: Array1<i64>,
+    style_vector: Array1<f32>,
+    sdp_ratio: f32,
+    length_scale: f32,
+    noise_scale: f32,
+    noise_scale_w: f32,
+) -> Result<Array3<f32>> {
+    let bert_ori = bert_ori.insert_axis(Axis(0));
+    let bert_ori = bert_ori.as_standard_layout();
+    let bert = ort::value::TensorRef::from_array_view(&bert_ori)?;
+    let mut x_tst_lengths = array![x_tst.shape()[0] as i64];
+    let x_tst_lengths = ort::value::TensorRef::from_array_view(&mut x_tst_lengths)?;
+    let mut x_tst = x_tst.insert_axis(Axis(0));
+    let x_tst = ort::value::TensorRef::from_array_view(&mut x_tst)?;
+    let mut lang_ids = lang_ids.insert_axis(Axis(0));
+    let lang_ids = ort::value::TensorRef::from_array_view(&mut lang_ids)?;
+    let mut tones = tones.insert_axis(Axis(0));
+    let tones = ort::value::TensorRef::from_array_view(&mut tones)?;
+    let mut style_vector = style_vector.insert_axis(Axis(0));
+    let style_vector = ort::value::TensorRef::from_array_view(&mut style_vector)?;
+    let sid = ort::value::TensorRef::from_array_view(&mut spk_ids)?;
+    let sdp_ratio = vec![sdp_ratio];
+    let sdp_ratio = ort::value::TensorRef::from_array_view((vec![1_i64], sdp_ratio.as_slice()))?;
+    let length_scale = vec![length_scale];
+    let length_scale =
+        ort::value::TensorRef::from_array_view((vec![1_i64], length_scale.as_slice()))?;
+    let noise_scale = vec![noise_scale];
+    let noise_scale =
+        ort::value::TensorRef::from_array_view((vec![1_i64], noise_scale.as_slice()))?;
+    let noise_scale_w = vec![noise_scale_w];
+    let noise_scale_w =
+        ort::value::TensorRef::from_array_view((vec![1_i64], noise_scale_w.as_slice()))?;
+    let outputs = session.run(ort::inputs! {
+        "x_tst" =>  x_tst,
+        "x_tst_lengths" => x_tst_lengths,
+        "sid" => sid,
+        "tones" => tones,
+        "language" => lang_ids,
+        "bert" => bert,
+        "style_vec" => style_vector,
+        "sdp_ratio" => sdp_ratio,
+        "length_scale" => length_scale,
+        "noise_scale" => noise_scale,
+        "noise_scale_w" => noise_scale_w,
+    })?;
+
+    let audio_array = outputs["output"]
+        .try_extract_tensor::<f32>()?
+        .into_dimensionality::<Ix3>()?
+        .to_owned();
+
+    Ok(audio_array)
+}

crates/sbv2_core/src/tts.rs

@@ -1,7 +1,13 @@
 use crate::error::{Error, Result};
 use crate::{jtalk, model, style, tokenizer, tts_util};
+#[cfg(feature = "aivmx")]
+use base64::prelude::{Engine as _, BASE64_STANDARD};
+#[cfg(feature = "aivmx")]
+use ndarray::ShapeBuilder;
 use ndarray::{concatenate, Array1, Array2, Array3, Axis};
-use ort::Session;
+use ort::session::Session;
+#[cfg(feature = "aivmx")]
+use std::io::Cursor;
 use tokenizers::Tokenizer;
 
 #[derive(PartialEq, Eq, Clone)]
@@ -69,6 +75,53 @@ impl TTSModelHolder {
         self.models.iter().map(|m| m.ident.to_string()).collect()
     }
 
+    #[cfg(feature = "aivmx")]
+    pub fn load_aivmx<I: Into<TTSIdent>, P: AsRef<[u8]>>(
+        &mut self,
+        ident: I,
+        aivmx_bytes: P,
+    ) -> Result<()> {
+        let ident = ident.into();
+        if self.find_model(ident.clone()).is_err() {
+            let mut load = true;
+            if let Some(max) = self.max_loaded_models {
+                if self.models.iter().filter(|x| x.vits2.is_some()).count() >= max {
+                    load = false;
+                }
+            }
+            let model = model::load_model(&aivmx_bytes, false)?;
+            let metadata = model.metadata()?;
+            if let Some(aivm_style_vectors) = metadata.custom("aivm_style_vectors")? {
+                let aivm_style_vectors = BASE64_STANDARD.decode(aivm_style_vectors)?;
+                let style_vectors = Cursor::new(&aivm_style_vectors);
+                let reader = npyz::NpyFile::new(style_vectors)?;
+                let style_vectors = {
+                    let shape = reader.shape().to_vec();
+                    let order = reader.order();
+                    let data = reader.into_vec::<f32>()?;
+                    let shape = match shape[..] {
+                        [i1, i2] => [i1 as usize, i2 as usize],
+                        _ => panic!("expected 2D array"),
+                    };
+                    let true_shape = shape.set_f(order == npyz::Order::Fortran);
+                    ndarray::Array2::from_shape_vec(true_shape, data)?
+                };
+                drop(metadata);
+                self.models.push(TTSModel {
+                    vits2: if load { Some(model) } else { None },
+                    bytes: if self.max_loaded_models.is_some() {
+                        Some(aivmx_bytes.as_ref().to_vec())
+                    } else {
+                        None
+                    },
+                    ident,
+                    style_vectors,
+                })
+            }
+        }
+        Ok(())
+    }
+
     /// Load a .sbv2 file binary
     ///
     /// # Examples
@@ -147,7 +200,7 @@ impl TTSModelHolder {
     /// This function is for low-level usage, use `easy_synthesize` for high-level usage.
     #[allow(clippy::type_complexity)]
     pub fn parse_text(
-        &self,
+        &mut self,
         text: &str,
     ) -> Result<(Array2<f32>, Array1<i64>, Array1<i64>, Array1<i64>)> {
         crate::tts_util::parse_text_blocking(
@@ -155,15 +208,15 @@ impl TTSModelHolder {
             &self.jtalk,
             &self.tokenizer,
             |token_ids, attention_masks| {
-                crate::bert::predict(&self.bert, token_ids, attention_masks)
+                crate::bert::predict(&mut self.bert, token_ids, attention_masks)
             },
         )
     }
 
-    fn find_model<I: Into<TTSIdent>>(&self, ident: I) -> Result<&TTSModel> {
+    fn find_model<I: Into<TTSIdent>>(&mut self, ident: I) -> Result<&mut TTSModel> {
         let ident = ident.into();
         self.models
-            .iter()
+            .iter_mut()
             .find(|m| m.ident == ident)
             .ok_or(Error::ModelNotFoundError(ident.to_string()))
     }
@@ -209,7 +262,7 @@ impl TTSModelHolder {
     /// # Note
     /// This function is for low-level usage, use `easy_synthesize` for high-level usage.
     pub fn get_style_vector<I: Into<TTSIdent>>(
-        &self,
+        &mut self,
         ident: I,
         style_id: i32,
         weight: f32,
@@ -233,11 +286,6 @@ impl TTSModelHolder {
         options: SynthesizeOptions,
     ) -> Result<Vec<u8>> {
         self.find_and_load_model(ident)?;
-        let vits2 = &self
-            .find_model(ident)?
-            .vits2
-            .as_ref()
-            .ok_or(Error::ModelNotFoundError(ident.into().to_string()))?;
         let style_vector = self.get_style_vector(ident, style_id, options.style_weight)?;
         let audio_array = if options.split_sentences {
             let texts: Vec<&str> = text.split('\n').collect();
@@ -247,6 +295,12 @@ impl TTSModelHolder {
                     continue;
                 }
                 let (bert_ori, phones, tones, lang_ids) = self.parse_text(t)?;
+
+                let vits2 = self
+                    .find_model(ident)?
+                    .vits2
+                    .as_mut()
+                    .ok_or(Error::ModelNotFoundError(ident.into().to_string()))?;
                 let audio = model::synthesize(
                     vits2,
                     bert_ori.to_owned(),
@@ -257,6 +311,8 @@ impl TTSModelHolder {
                     style_vector.clone(),
                     options.sdp_ratio,
                     options.length_scale,
+                    0.677,
+                    0.8,
                 )?;
                 audios.push(audio.clone());
                 if i != texts.len() - 1 {
@@ -269,6 +325,12 @@ impl TTSModelHolder {
             )?
         } else {
             let (bert_ori, phones, tones, lang_ids) = self.parse_text(text)?;
+
+            let vits2 = self
+                .find_model(ident)?
+                .vits2
+                .as_mut()
+                .ok_or(Error::ModelNotFoundError(ident.into().to_string()))?;
             model::synthesize(
                 vits2,
                 bert_ori.to_owned(),
@@ -279,6 +341,8 @@ impl TTSModelHolder {
                 style_vector,
                 options.sdp_ratio,
                 options.length_scale,
+                0.677,
+                0.8,
             )?
         };
         tts_util::array_to_vec(audio_array)

crates/sbv2_wasm/Cargo.toml

@@ -1,7 +1,12 @@
 [package]
 name = "sbv2_wasm"
-version = "0.1.0"
-edition = "2021"
+version.workspace = true
+edition.workspace = true
+description.workspace = true
+readme.workspace = true
+repository.workspace = true
+documentation.workspace = true
+license.workspace = true
 
 [lib]
 crate-type = ["cdylib", "rlib"]
@@ -13,7 +18,3 @@ once_cell.workspace = true
 js-sys = "0.3.70"
 ndarray.workspace = true
 wasm-bindgen-futures = "0.4.43"
-
-[profile.release]
-lto = true
-opt-level = "s"

crates/sbv2_wasm/build.sh

@@ -0,0 +1,8 @@
+#!/bin/sh
+wasm-pack build --target web ./crates/sbv2_wasm --release
+wasm-opt -O3 -o ./crates/sbv2_wasm/pkg/sbv2_wasm_bg.wasm ./crates/sbv2_wasm/pkg/sbv2_wasm_bg.wasm
+wasm-strip ./crates/sbv2_wasm/pkg/sbv2_wasm_bg.wasm
+mkdir -p ./crates/sbv2_wasm/dist
+cp ./crates/sbv2_wasm/pkg/sbv2_wasm_bg.wasm ./crates/sbv2_wasm/dist/sbv2_wasm_bg.wasm
+cd ./crates/sbv2_wasm
+pnpm build

crates/sbv2_wasm/example.js

@@ -3,9 +3,12 @@ import fs from "node:fs/promises";
 
 ModelHolder.globalInit(await fs.readFile("./dist/sbv2_wasm_bg.wasm"));
 const holder = await ModelHolder.create(
-	(await fs.readFile("../models/tokenizer.json")).toString("utf-8"),
-	await fs.readFile("../models/deberta.onnx"),
+	(await fs.readFile("../../models/tokenizer.json")).toString("utf-8"),
+	await fs.readFile("../../models/deberta.onnx"),
+);
+await holder.load(
+	"tsukuyomi",
+	await fs.readFile("../../models/tsukuyomi.sbv2"),
 );
-await holder.load("tsukuyomi", await fs.readFile("../models/iroha2.sbv2"));
 await fs.writeFile("out.wav", await holder.synthesize("tsukuyomi", "おはよう"));
 holder.unload("tsukuyomi");

crates/sbv2_wasm/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "sbv2",
-	"version": "0.1.1",
+	"version": "0.2.0-alpha6",
 	"description": "Style Bert VITS2 wasm",
 	"main": "dist/index.js",
 	"types": "dist/index.d.ts",
@@ -13,13 +13,13 @@
 	"author": "tuna2134",
 	"license": "MIT",
 	"devDependencies": {
-		"@biomejs/biome": "^1.9.2",
-		"@types/node": "^22.7.4",
-		"esbuild": "^0.24.0",
-		"typescript": "^5.6.2"
+		"@biomejs/biome": "^1.9.4",
+		"@types/node": "^22.13.5",
+		"esbuild": "^0.25.0",
+		"typescript": "^5.7.3"
 	},
 	"dependencies": {
-		"onnxruntime-web": "^1.19.2"
+		"onnxruntime-web": "^1.20.1"
 	},
-	"files": ["dist/*", "package.json", "README.md"]
+	"files": ["dist/*", "package.json", "README.md", "pkg/*.ts", "pkg/*.js"]
 }

crates/sbv2_wasm/pnpm-lock.yaml

@@ -9,21 +9,21 @@ importers:
   .:
     dependencies:
       onnxruntime-web:
-        specifier: ^1.19.2
-        version: 1.20.0
+        specifier: ^1.20.1
+        version: 1.20.1
     devDependencies:
       '@biomejs/biome':
-        specifier: ^1.9.2
+        specifier: ^1.9.4
         version: 1.9.4
       '@types/node':
-        specifier: ^22.7.4
-        version: 22.8.0
+        specifier: ^22.13.5
+        version: 22.13.5
       esbuild:
-        specifier: ^0.24.0
-        version: 0.24.0
+        specifier: ^0.25.0
+        version: 0.25.0
       typescript:
-        specifier: ^5.6.2
-        version: 5.6.3
+        specifier: ^5.7.3
+        version: 5.7.3
 
 packages:
 
@@ -80,146 +80,152 @@ packages:
     cpu: [x64]
     os: [win32]
 
-  '@esbuild/aix-ppc64@0.24.0':
-    resolution: {integrity: sha512-WtKdFM7ls47zkKHFVzMz8opM7LkcsIp9amDUBIAWirg70RM71WRSjdILPsY5Uv1D42ZpUfaPILDlfactHgsRkw==}
+  '@esbuild/aix-ppc64@0.25.0':
+    resolution: {integrity: sha512-O7vun9Sf8DFjH2UtqK8Ku3LkquL9SZL8OLY1T5NZkA34+wG3OQF7cl4Ql8vdNzM6fzBbYfLaiRLIOZ+2FOCgBQ==}
     engines: {node: '>=18'}
     cpu: [ppc64]
     os: [aix]
 
-  '@esbuild/android-arm64@0.24.0':
-    resolution: {integrity: sha512-Vsm497xFM7tTIPYK9bNTYJyF/lsP590Qc1WxJdlB6ljCbdZKU9SY8i7+Iin4kyhV/KV5J2rOKsBQbB77Ab7L/w==}
+  '@esbuild/android-arm64@0.25.0':
+    resolution: {integrity: sha512-grvv8WncGjDSyUBjN9yHXNt+cq0snxXbDxy5pJtzMKGmmpPxeAmAhWxXI+01lU5rwZomDgD3kJwulEnhTRUd6g==}
     engines: {node: '>=18'}
     cpu: [arm64]
     os: [android]
 
-  '@esbuild/android-arm@0.24.0':
-    resolution: {integrity: sha512-arAtTPo76fJ/ICkXWetLCc9EwEHKaeya4vMrReVlEIUCAUncH7M4bhMQ+M9Vf+FFOZJdTNMXNBrWwW+OXWpSew==}
+  '@esbuild/android-arm@0.25.0':
+    resolution: {integrity: sha512-PTyWCYYiU0+1eJKmw21lWtC+d08JDZPQ5g+kFyxP0V+es6VPPSUhM6zk8iImp2jbV6GwjX4pap0JFbUQN65X1g==}
     engines: {node: '>=18'}
     cpu: [arm]
     os: [android]
 
-  '@esbuild/android-x64@0.24.0':
-    resolution: {integrity: sha512-t8GrvnFkiIY7pa7mMgJd7p8p8qqYIz1NYiAoKc75Zyv73L3DZW++oYMSHPRarcotTKuSs6m3hTOa5CKHaS02TQ==}
+  '@esbuild/android-x64@0.25.0':
+    resolution: {integrity: sha512-m/ix7SfKG5buCnxasr52+LI78SQ+wgdENi9CqyCXwjVR2X4Jkz+BpC3le3AoBPYTC9NHklwngVXvbJ9/Akhrfg==}
     engines: {node: '>=18'}
     cpu: [x64]
     os: [android]
 
-  '@esbuild/darwin-arm64@0.24.0':
-    resolution: {integrity: sha512-CKyDpRbK1hXwv79soeTJNHb5EiG6ct3efd/FTPdzOWdbZZfGhpbcqIpiD0+vwmpu0wTIL97ZRPZu8vUt46nBSw==}
+  '@esbuild/darwin-arm64@0.25.0':
+    resolution: {integrity: sha512-mVwdUb5SRkPayVadIOI78K7aAnPamoeFR2bT5nszFUZ9P8UpK4ratOdYbZZXYSqPKMHfS1wdHCJk1P1EZpRdvw==}
     engines: {node: '>=18'}
     cpu: [arm64]
     os: [darwin]
 
-  '@esbuild/darwin-x64@0.24.0':
-    resolution: {integrity: sha512-rgtz6flkVkh58od4PwTRqxbKH9cOjaXCMZgWD905JOzjFKW+7EiUObfd/Kav+A6Gyud6WZk9w+xu6QLytdi2OA==}
+  '@esbuild/darwin-x64@0.25.0':
+    resolution: {integrity: sha512-DgDaYsPWFTS4S3nWpFcMn/33ZZwAAeAFKNHNa1QN0rI4pUjgqf0f7ONmXf6d22tqTY+H9FNdgeaAa+YIFUn2Rg==}
     engines: {node: '>=18'}
     cpu: [x64]
     os: [darwin]
 
-  '@esbuild/freebsd-arm64@0.24.0':
-    resolution: {integrity: sha512-6Mtdq5nHggwfDNLAHkPlyLBpE5L6hwsuXZX8XNmHno9JuL2+bg2BX5tRkwjyfn6sKbxZTq68suOjgWqCicvPXA==}
+  '@esbuild/freebsd-arm64@0.25.0':
+    resolution: {integrity: sha512-VN4ocxy6dxefN1MepBx/iD1dH5K8qNtNe227I0mnTRjry8tj5MRk4zprLEdG8WPyAPb93/e4pSgi1SoHdgOa4w==}
     engines: {node: '>=18'}
     cpu: [arm64]
     os: [freebsd]
 
-  '@esbuild/freebsd-x64@0.24.0':
-    resolution: {integrity: sha512-D3H+xh3/zphoX8ck4S2RxKR6gHlHDXXzOf6f/9dbFt/NRBDIE33+cVa49Kil4WUjxMGW0ZIYBYtaGCa2+OsQwQ==}
+  '@esbuild/freebsd-x64@0.25.0':
+    resolution: {integrity: sha512-mrSgt7lCh07FY+hDD1TxiTyIHyttn6vnjesnPoVDNmDfOmggTLXRv8Id5fNZey1gl/V2dyVK1VXXqVsQIiAk+A==}
     engines: {node: '>=18'}
     cpu: [x64]
     os: [freebsd]
 
-  '@esbuild/linux-arm64@0.24.0':
-    resolution: {integrity: sha512-TDijPXTOeE3eaMkRYpcy3LarIg13dS9wWHRdwYRnzlwlA370rNdZqbcp0WTyyV/k2zSxfko52+C7jU5F9Tfj1g==}
+  '@esbuild/linux-arm64@0.25.0':
+    resolution: {integrity: sha512-9QAQjTWNDM/Vk2bgBl17yWuZxZNQIF0OUUuPZRKoDtqF2k4EtYbpyiG5/Dk7nqeK6kIJWPYldkOcBqjXjrUlmg==}
     engines: {node: '>=18'}
     cpu: [arm64]
     os: [linux]
 
-  '@esbuild/linux-arm@0.24.0':
-    resolution: {integrity: sha512-gJKIi2IjRo5G6Glxb8d3DzYXlxdEj2NlkixPsqePSZMhLudqPhtZ4BUrpIuTjJYXxvF9njql+vRjB2oaC9XpBw==}
+  '@esbuild/linux-arm@0.25.0':
+    resolution: {integrity: sha512-vkB3IYj2IDo3g9xX7HqhPYxVkNQe8qTK55fraQyTzTX/fxaDtXiEnavv9geOsonh2Fd2RMB+i5cbhu2zMNWJwg==}
     engines: {node: '>=18'}
     cpu: [arm]
     os: [linux]
 
-  '@esbuild/linux-ia32@0.24.0':
-    resolution: {integrity: sha512-K40ip1LAcA0byL05TbCQ4yJ4swvnbzHscRmUilrmP9Am7//0UjPreh4lpYzvThT2Quw66MhjG//20mrufm40mA==}
+  '@esbuild/linux-ia32@0.25.0':
+    resolution: {integrity: sha512-43ET5bHbphBegyeqLb7I1eYn2P/JYGNmzzdidq/w0T8E2SsYL1U6un2NFROFRg1JZLTzdCoRomg8Rvf9M6W6Gg==}
     engines: {node: '>=18'}
     cpu: [ia32]
     os: [linux]
 
-  '@esbuild/linux-loong64@0.24.0':
-    resolution: {integrity: sha512-0mswrYP/9ai+CU0BzBfPMZ8RVm3RGAN/lmOMgW4aFUSOQBjA31UP8Mr6DDhWSuMwj7jaWOT0p0WoZ6jeHhrD7g==}
+  '@esbuild/linux-loong64@0.25.0':
+    resolution: {integrity: sha512-fC95c/xyNFueMhClxJmeRIj2yrSMdDfmqJnyOY4ZqsALkDrrKJfIg5NTMSzVBr5YW1jf+l7/cndBfP3MSDpoHw==}
     engines: {node: '>=18'}
     cpu: [loong64]
     os: [linux]
 
-  '@esbuild/linux-mips64el@0.24.0':
-    resolution: {integrity: sha512-hIKvXm0/3w/5+RDtCJeXqMZGkI2s4oMUGj3/jM0QzhgIASWrGO5/RlzAzm5nNh/awHE0A19h/CvHQe6FaBNrRA==}
+  '@esbuild/linux-mips64el@0.25.0':
+    resolution: {integrity: sha512-nkAMFju7KDW73T1DdH7glcyIptm95a7Le8irTQNO/qtkoyypZAnjchQgooFUDQhNAy4iu08N79W4T4pMBwhPwQ==}
     engines: {node: '>=18'}
     cpu: [mips64el]
     os: [linux]
 
-  '@esbuild/linux-ppc64@0.24.0':
-    resolution: {integrity: sha512-HcZh5BNq0aC52UoocJxaKORfFODWXZxtBaaZNuN3PUX3MoDsChsZqopzi5UupRhPHSEHotoiptqikjN/B77mYQ==}
+  '@esbuild/linux-ppc64@0.25.0':
+    resolution: {integrity: sha512-NhyOejdhRGS8Iwv+KKR2zTq2PpysF9XqY+Zk77vQHqNbo/PwZCzB5/h7VGuREZm1fixhs4Q/qWRSi5zmAiO4Fw==}
     engines: {node: '>=18'}
     cpu: [ppc64]
     os: [linux]
 
-  '@esbuild/linux-riscv64@0.24.0':
-    resolution: {integrity: sha512-bEh7dMn/h3QxeR2KTy1DUszQjUrIHPZKyO6aN1X4BCnhfYhuQqedHaa5MxSQA/06j3GpiIlFGSsy1c7Gf9padw==}
+  '@esbuild/linux-riscv64@0.25.0':
+    resolution: {integrity: sha512-5S/rbP5OY+GHLC5qXp1y/Mx//e92L1YDqkiBbO9TQOvuFXM+iDqUNG5XopAnXoRH3FjIUDkeGcY1cgNvnXp/kA==}
     engines: {node: '>=18'}
     cpu: [riscv64]
     os: [linux]
 
-  '@esbuild/linux-s390x@0.24.0':
-    resolution: {integrity: sha512-ZcQ6+qRkw1UcZGPyrCiHHkmBaj9SiCD8Oqd556HldP+QlpUIe2Wgn3ehQGVoPOvZvtHm8HPx+bH20c9pvbkX3g==}
+  '@esbuild/linux-s390x@0.25.0':
+    resolution: {integrity: sha512-XM2BFsEBz0Fw37V0zU4CXfcfuACMrppsMFKdYY2WuTS3yi8O1nFOhil/xhKTmE1nPmVyvQJjJivgDT+xh8pXJA==}
     engines: {node: '>=18'}
     cpu: [s390x]
     os: [linux]
 
-  '@esbuild/linux-x64@0.24.0':
-    resolution: {integrity: sha512-vbutsFqQ+foy3wSSbmjBXXIJ6PL3scghJoM8zCL142cGaZKAdCZHyf+Bpu/MmX9zT9Q0zFBVKb36Ma5Fzfa8xA==}
+  '@esbuild/linux-x64@0.25.0':
+    resolution: {integrity: sha512-9yl91rHw/cpwMCNytUDxwj2XjFpxML0y9HAOH9pNVQDpQrBxHy01Dx+vaMu0N1CKa/RzBD2hB4u//nfc+Sd3Cw==}
     engines: {node: '>=18'}
     cpu: [x64]
     os: [linux]
 
-  '@esbuild/netbsd-x64@0.24.0':
-    resolution: {integrity: sha512-hjQ0R/ulkO8fCYFsG0FZoH+pWgTTDreqpqY7UnQntnaKv95uP5iW3+dChxnx7C3trQQU40S+OgWhUVwCjVFLvg==}
+  '@esbuild/netbsd-arm64@0.25.0':
+    resolution: {integrity: sha512-RuG4PSMPFfrkH6UwCAqBzauBWTygTvb1nxWasEJooGSJ/NwRw7b2HOwyRTQIU97Hq37l3npXoZGYMy3b3xYvPw==}
+    engines: {node: '>=18'}
+    cpu: [arm64]
+    os: [netbsd]
+
+  '@esbuild/netbsd-x64@0.25.0':
+    resolution: {integrity: sha512-jl+qisSB5jk01N5f7sPCsBENCOlPiS/xptD5yxOx2oqQfyourJwIKLRA2yqWdifj3owQZCL2sn6o08dBzZGQzA==}
     engines: {node: '>=18'}
     cpu: [x64]
     os: [netbsd]
 
-  '@esbuild/openbsd-arm64@0.24.0':
-    resolution: {integrity: sha512-MD9uzzkPQbYehwcN583yx3Tu5M8EIoTD+tUgKF982WYL9Pf5rKy9ltgD0eUgs8pvKnmizxjXZyLt0z6DC3rRXg==}
+  '@esbuild/openbsd-arm64@0.25.0':
+    resolution: {integrity: sha512-21sUNbq2r84YE+SJDfaQRvdgznTD8Xc0oc3p3iW/a1EVWeNj/SdUCbm5U0itZPQYRuRTW20fPMWMpcrciH2EJw==}
     engines: {node: '>=18'}
     cpu: [arm64]
     os: [openbsd]
 
-  '@esbuild/openbsd-x64@0.24.0':
-    resolution: {integrity: sha512-4ir0aY1NGUhIC1hdoCzr1+5b43mw99uNwVzhIq1OY3QcEwPDO3B7WNXBzaKY5Nsf1+N11i1eOfFcq+D/gOS15Q==}
+  '@esbuild/openbsd-x64@0.25.0':
+    resolution: {integrity: sha512-2gwwriSMPcCFRlPlKx3zLQhfN/2WjJ2NSlg5TKLQOJdV0mSxIcYNTMhk3H3ulL/cak+Xj0lY1Ym9ysDV1igceg==}
     engines: {node: '>=18'}
     cpu: [x64]
     os: [openbsd]
 
-  '@esbuild/sunos-x64@0.24.0':
-    resolution: {integrity: sha512-jVzdzsbM5xrotH+W5f1s+JtUy1UWgjU0Cf4wMvffTB8m6wP5/kx0KiaLHlbJO+dMgtxKV8RQ/JvtlFcdZ1zCPA==}
+  '@esbuild/sunos-x64@0.25.0':
+    resolution: {integrity: sha512-bxI7ThgLzPrPz484/S9jLlvUAHYMzy6I0XiU1ZMeAEOBcS0VePBFxh1JjTQt3Xiat5b6Oh4x7UC7IwKQKIJRIg==}
     engines: {node: '>=18'}
     cpu: [x64]
     os: [sunos]
 
-  '@esbuild/win32-arm64@0.24.0':
-    resolution: {integrity: sha512-iKc8GAslzRpBytO2/aN3d2yb2z8XTVfNV0PjGlCxKo5SgWmNXx82I/Q3aG1tFfS+A2igVCY97TJ8tnYwpUWLCA==}
+  '@esbuild/win32-arm64@0.25.0':
+    resolution: {integrity: sha512-ZUAc2YK6JW89xTbXvftxdnYy3m4iHIkDtK3CLce8wg8M2L+YZhIvO1DKpxrd0Yr59AeNNkTiic9YLf6FTtXWMw==}
     engines: {node: '>=18'}
     cpu: [arm64]
     os: [win32]
 
-  '@esbuild/win32-ia32@0.24.0':
-    resolution: {integrity: sha512-vQW36KZolfIudCcTnaTpmLQ24Ha1RjygBo39/aLkM2kmjkWmZGEJ5Gn9l5/7tzXA42QGIoWbICfg6KLLkIw6yw==}
+  '@esbuild/win32-ia32@0.25.0':
+    resolution: {integrity: sha512-eSNxISBu8XweVEWG31/JzjkIGbGIJN/TrRoiSVZwZ6pkC6VX4Im/WV2cz559/TXLcYbcrDN8JtKgd9DJVIo8GA==}
     engines: {node: '>=18'}
     cpu: [ia32]
     os: [win32]
 
-  '@esbuild/win32-x64@0.24.0':
-    resolution: {integrity: sha512-7IAFPrjSQIJrGsK6flwg7NFmwBoSTyF3rl7If0hNUFQU4ilTsEPL6GuMuU9BfIWVVGuRnuIidkSMC+c0Otu8IA==}
+  '@esbuild/win32-x64@0.25.0':
+    resolution: {integrity: sha512-ZENoHJBxA20C2zFzh6AI4fT6RraMzjYw4xKWemRTRmRVtN9c5DcH9r/f2ihEkMjOW5eGgrwCslG/+Y/3bL+DHQ==}
     engines: {node: '>=18'}
     cpu: [x64]
     os: [win32]
@@ -254,11 +260,11 @@ packages:
   '@protobufjs/utf8@1.1.0':
     resolution: {integrity: sha512-Vvn3zZrhQZkkBE8LSuW3em98c0FwgO4nxzv6OdSxPKJIEKY2bGbHn+mhGIPerzI4twdxaP8/0+06HBpwf345Lw==}
 
-  '@types/node@22.8.0':
-    resolution: {integrity: sha512-84rafSBHC/z1i1E3p0cJwKA+CfYDNSXX9WSZBRopjIzLET8oNt6ht2tei4C7izwDeEiLLfdeSVBv1egOH916hg==}
+  '@types/node@22.13.5':
+    resolution: {integrity: sha512-+lTU0PxZXn0Dr1NBtC7Y8cR21AJr87dLLU953CWA6pMxxv/UDc7jYAY90upcrie1nRcD6XNG5HOYEDtgW5TxAg==}
 
-  esbuild@0.24.0:
-    resolution: {integrity: sha512-FuLPevChGDshgSicjisSooU0cemp/sGXR841D5LHMB7mTVOmsEHcAxaH3irL53+8YDIeVNQEySh4DaYU/iuPqQ==}
+  esbuild@0.25.0:
+    resolution: {integrity: sha512-BXq5mqc8ltbaN34cDqWuYKyNhX8D/Z0J1xdtdQ8UcIIIyJyz+ZMKUt58tF3SrZ85jcfN/PZYhjR5uDQAYNVbuw==}
     engines: {node: '>=18'}
     hasBin: true
 
@@ -268,14 +274,14 @@ packages:
   guid-typescript@1.0.9:
     resolution: {integrity: sha512-Y8T4vYhEfwJOTbouREvG+3XDsjr8E3kIr7uf+JZ0BYloFsttiHU0WfvANVsR7TxNUJa/WpCnw/Ino/p+DeBhBQ==}
 
-  long@5.2.3:
-    resolution: {integrity: sha512-lcHwpNoggQTObv5apGNCTdJrO69eHOZMi4BNC+rTLER8iHAqGrUVeLh/irVIM7zTw2bOXA8T6uNPeujwOLg/2Q==}
+  long@5.3.1:
+    resolution: {integrity: sha512-ka87Jz3gcx/I7Hal94xaN2tZEOPoUOEVftkQqZx2EeQRN7LGdfLlI3FvZ+7WDplm+vK2Urx9ULrvSowtdCieng==}
 
-  onnxruntime-common@1.20.0:
-    resolution: {integrity: sha512-9ehS4ul5fBszIcHhfxuDgk45lO+Fqrxmrgwk1Pxb1JRvbQiCB/v9Royv95SRCWHktLMviqNjBsEd/biJhd39cg==}
+  onnxruntime-common@1.20.1:
+    resolution: {integrity: sha512-YiU0s0IzYYC+gWvqD1HzLc46Du1sXpSiwzKb63PACIJr6LfL27VsXSXQvt68EzD3V0D5Bc0vyJTjmMxp0ylQiw==}
 
-  onnxruntime-web@1.20.0:
-    resolution: {integrity: sha512-IoUf8dqHFJLV4DUSz+Ok+xxyN6cQk57gb20m6PZE5gag3QXuvegYMq9dG8t/QF4JjTKIwvfvnr16ouzCCB9IMA==}
+  onnxruntime-web@1.20.1:
+    resolution: {integrity: sha512-TePF6XVpLL1rWVMIl5Y9ACBQcyCNFThZON/jgElNd9Txb73CIEGlklhYR3UEr1cp5r0rbGI6nDwwrs79g7WjoA==}
 
   platform@1.3.6:
     resolution: {integrity: sha512-fnWVljUchTro6RiCFvCXBbNhJc2NijN7oIQxbwsyL0buWJPG85v81ehlHI9fXrJsMNgTofEoWIQeClKpgxFLrg==}
@@ -284,13 +290,13 @@ packages:
     resolution: {integrity: sha512-mRUWCc3KUU4w1jU8sGxICXH/gNS94DvI1gxqDvBzhj1JpcsimQkYiOJfwsPUykUI5ZaspFbSgmBLER8IrQ3tqw==}
     engines: {node: '>=12.0.0'}
 
-  typescript@5.6.3:
-    resolution: {integrity: sha512-hjcS1mhfuyi4WW8IWtjP7brDrG2cuDZukyrYrSauoXGNgx0S7zceP07adYkJycEr56BOUTNPzbInooiN3fn1qw==}
+  typescript@5.7.3:
+    resolution: {integrity: sha512-84MVSjMEHP+FQRPy3pX9sTVV/INIex71s9TL2Gm5FG/WG1SqXeKyZ0k7/blY/4FdOzI12CBy1vGc4og/eus0fw==}
     engines: {node: '>=14.17'}
     hasBin: true
 
-  undici-types@6.19.8:
-    resolution: {integrity: sha512-ve2KP6f/JnbPBFyobGHuerC9g1FYGn/F8n1LWTwNxCEzd6IfqTwUQcNXgEtmmQ6DlRrC1hrSrBnCZPokRrDHjw==}
+  undici-types@6.20.0:
+    resolution: {integrity: sha512-Ny6QZ2Nju20vw1SRHe3d9jVu6gJ+4e3+MMpqu7pqE5HT6WsTSlce++GQmK5UXS8mzV8DSYHrQH+Xrf2jVcuKNg==}
 
 snapshots:
 
@@ -329,76 +335,79 @@ snapshots:
   '@biomejs/cli-win32-x64@1.9.4':
     optional: true
 
-  '@esbuild/aix-ppc64@0.24.0':
+  '@esbuild/aix-ppc64@0.25.0':
     optional: true
 
-  '@esbuild/android-arm64@0.24.0':
+  '@esbuild/android-arm64@0.25.0':
     optional: true
 
-  '@esbuild/android-arm@0.24.0':
+  '@esbuild/android-arm@0.25.0':
     optional: true
 
-  '@esbuild/android-x64@0.24.0':
+  '@esbuild/android-x64@0.25.0':
     optional: true
 
-  '@esbuild/darwin-arm64@0.24.0':
+  '@esbuild/darwin-arm64@0.25.0':
     optional: true
 
-  '@esbuild/darwin-x64@0.24.0':
+  '@esbuild/darwin-x64@0.25.0':
     optional: true
 
-  '@esbuild/freebsd-arm64@0.24.0':
+  '@esbuild/freebsd-arm64@0.25.0':
     optional: true
 
-  '@esbuild/freebsd-x64@0.24.0':
+  '@esbuild/freebsd-x64@0.25.0':
     optional: true
 
-  '@esbuild/linux-arm64@0.24.0':
+  '@esbuild/linux-arm64@0.25.0':
     optional: true
 
-  '@esbuild/linux-arm@0.24.0':
+  '@esbuild/linux-arm@0.25.0':
     optional: true
 
-  '@esbuild/linux-ia32@0.24.0':
+  '@esbuild/linux-ia32@0.25.0':
     optional: true
 
-  '@esbuild/linux-loong64@0.24.0':
+  '@esbuild/linux-loong64@0.25.0':
     optional: true
 
-  '@esbuild/linux-mips64el@0.24.0':
+  '@esbuild/linux-mips64el@0.25.0':
     optional: true
 
-  '@esbuild/linux-ppc64@0.24.0':
+  '@esbuild/linux-ppc64@0.25.0':
     optional: true
 
-  '@esbuild/linux-riscv64@0.24.0':
+  '@esbuild/linux-riscv64@0.25.0':
     optional: true
 
-  '@esbuild/linux-s390x@0.24.0':
+  '@esbuild/linux-s390x@0.25.0':
     optional: true
 
-  '@esbuild/linux-x64@0.24.0':
+  '@esbuild/linux-x64@0.25.0':
     optional: true
 
-  '@esbuild/netbsd-x64@0.24.0':
+  '@esbuild/netbsd-arm64@0.25.0':
     optional: true
 
-  '@esbuild/openbsd-arm64@0.24.0':
+  '@esbuild/netbsd-x64@0.25.0':
     optional: true
 
-  '@esbuild/openbsd-x64@0.24.0':
+  '@esbuild/openbsd-arm64@0.25.0':
     optional: true
 
-  '@esbuild/sunos-x64@0.24.0':
+  '@esbuild/openbsd-x64@0.25.0':
     optional: true
 
-  '@esbuild/win32-arm64@0.24.0':
+  '@esbuild/sunos-x64@0.25.0':
     optional: true
 
-  '@esbuild/win32-ia32@0.24.0':
+  '@esbuild/win32-arm64@0.25.0':
     optional: true
 
-  '@esbuild/win32-x64@0.24.0':
+  '@esbuild/win32-ia32@0.25.0':
+    optional: true
+
+  '@esbuild/win32-x64@0.25.0':
     optional: true
 
   '@protobufjs/aspromise@1.1.2': {}
@@ -424,51 +433,52 @@ snapshots:
 
   '@protobufjs/utf8@1.1.0': {}
 
-  '@types/node@22.8.0':
+  '@types/node@22.13.5':
     dependencies:
-      undici-types: 6.19.8
+      undici-types: 6.20.0
 
-  esbuild@0.24.0:
+  esbuild@0.25.0:
     optionalDependencies:
-      '@esbuild/aix-ppc64': 0.24.0
-      '@esbuild/android-arm': 0.24.0
-      '@esbuild/android-arm64': 0.24.0
-      '@esbuild/android-x64': 0.24.0
-      '@esbuild/darwin-arm64': 0.24.0
-      '@esbuild/darwin-x64': 0.24.0
-      '@esbuild/freebsd-arm64': 0.24.0
-      '@esbuild/freebsd-x64': 0.24.0
-      '@esbuild/linux-arm': 0.24.0
-      '@esbuild/linux-arm64': 0.24.0
-      '@esbuild/linux-ia32': 0.24.0
-      '@esbuild/linux-loong64': 0.24.0
-      '@esbuild/linux-mips64el': 0.24.0
-      '@esbuild/linux-ppc64': 0.24.0
-      '@esbuild/linux-riscv64': 0.24.0
-      '@esbuild/linux-s390x': 0.24.0
-      '@esbuild/linux-x64': 0.24.0
-      '@esbuild/netbsd-x64': 0.24.0
-      '@esbuild/openbsd-arm64': 0.24.0
-      '@esbuild/openbsd-x64': 0.24.0
-      '@esbuild/sunos-x64': 0.24.0
-      '@esbuild/win32-arm64': 0.24.0
-      '@esbuild/win32-ia32': 0.24.0
-      '@esbuild/win32-x64': 0.24.0
+      '@esbuild/aix-ppc64': 0.25.0
+      '@esbuild/android-arm': 0.25.0
+      '@esbuild/android-arm64': 0.25.0
+      '@esbuild/android-x64': 0.25.0
+      '@esbuild/darwin-arm64': 0.25.0
+      '@esbuild/darwin-x64': 0.25.0
+      '@esbuild/freebsd-arm64': 0.25.0
+      '@esbuild/freebsd-x64': 0.25.0
+      '@esbuild/linux-arm': 0.25.0
+      '@esbuild/linux-arm64': 0.25.0
+      '@esbuild/linux-ia32': 0.25.0
+      '@esbuild/linux-loong64': 0.25.0
+      '@esbuild/linux-mips64el': 0.25.0
+      '@esbuild/linux-ppc64': 0.25.0
+      '@esbuild/linux-riscv64': 0.25.0
+      '@esbuild/linux-s390x': 0.25.0
+      '@esbuild/linux-x64': 0.25.0
+      '@esbuild/netbsd-arm64': 0.25.0
+      '@esbuild/netbsd-x64': 0.25.0
+      '@esbuild/openbsd-arm64': 0.25.0
+      '@esbuild/openbsd-x64': 0.25.0
+      '@esbuild/sunos-x64': 0.25.0
+      '@esbuild/win32-arm64': 0.25.0
+      '@esbuild/win32-ia32': 0.25.0
+      '@esbuild/win32-x64': 0.25.0
 
   flatbuffers@1.12.0: {}
 
   guid-typescript@1.0.9: {}
 
-  long@5.2.3: {}
+  long@5.3.1: {}
 
-  onnxruntime-common@1.20.0: {}
+  onnxruntime-common@1.20.1: {}
 
-  onnxruntime-web@1.20.0:
+  onnxruntime-web@1.20.1:
     dependencies:
       flatbuffers: 1.12.0
       guid-typescript: 1.0.9
-      long: 5.2.3
-      onnxruntime-common: 1.20.0
+      long: 5.3.1
+      onnxruntime-common: 1.20.1
       platform: 1.3.6
       protobufjs: 7.4.0
 
@@ -486,9 +496,9 @@ snapshots:
       '@protobufjs/path': 1.1.2
       '@protobufjs/pool': 1.1.0
       '@protobufjs/utf8': 1.1.0
-      '@types/node': 22.8.0
-      long: 5.2.3
+      '@types/node': 22.13.5
+      long: 5.3.1
 
-  typescript@5.6.3: {}
+  typescript@5.7.3: {}
 
-  undici-types@6.19.8: {}
+  undici-types@6.20.0: {}

crates/sbv2_wasm/src-js/index.ts

@@ -74,6 +74,8 @@ export class ModelHolder {
 							style_vec: e,
 							sdp_ratio: new Tensor("float32", [f]),
 							length_scale: new Tensor("float32", [g]),
+							noise_scale: new Tensor("float32", [0.677]),
+							noise_scale_w: new Tensor("float32", [0.8]),
 						})
 					).output;
 					return [new Uint32Array(res.dims), await res.getData(true)];

sbv2_bindings/Cargo.toml

@@ -1,15 +0,0 @@
-[package]
-name = "sbv2_bindings"
-version = "0.2.0-alpha2"
-edition = "2021"
-
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
-[lib]
-name = "sbv2_bindings"
-crate-type = ["cdylib"]
-
-[dependencies]
-anyhow.workspace = true
-ndarray.workspace = true
-pyo3 = { version = "0.22.0", features = ["anyhow"] }
-sbv2_core = { version = "0.2.0-alpha2", path = "../sbv2_core" }

sbv2_core/src/model.rs

@@ -1,86 +0,0 @@
-use crate::error::Result;
-use ndarray::{array, Array1, Array2, Array3, Axis, Ix3};
-use ort::{GraphOptimizationLevel, Session};
-
-#[allow(clippy::vec_init_then_push, unused_variables)]
-pub fn load_model<P: AsRef<[u8]>>(model_file: P, bert: bool) -> Result<Session> {
-    let mut exp = Vec::new();
-    #[cfg(feature = "tensorrt")]
-    {
-        if bert {
-            exp.push(
-                ort::TensorRTExecutionProvider::default()
-                    .with_fp16(true)
-                    .with_profile_min_shapes("input_ids:1x1,attention_mask:1x1")
-                    .with_profile_max_shapes("input_ids:1x100,attention_mask:1x100")
-                    .with_profile_opt_shapes("input_ids:1x25,attention_mask:1x25")
-                    .build(),
-            );
-        }
-    }
-    #[cfg(feature = "cuda")]
-    {
-        #[allow(unused_mut)]
-        let mut cuda = ort::CUDAExecutionProvider::default()
-            .with_conv_algorithm_search(ort::CUDAExecutionProviderCuDNNConvAlgoSearch::Default);
-        #[cfg(feature = "cuda_tf32")]
-        {
-            cuda = cuda.with_tf32(true);
-        }
-        exp.push(cuda.build());
-    }
-    #[cfg(feature = "directml")]
-    {
-        exp.push(ort::DirectMLExecutionProvider::default().build());
-    }
-    #[cfg(feature = "coreml")]
-    {
-        exp.push(ort::CoreMLExecutionProvider::default().build());
-    }
-    exp.push(ort::CPUExecutionProvider::default().build());
-    Ok(Session::builder()?
-        .with_execution_providers(exp)?
-        .with_optimization_level(GraphOptimizationLevel::Level3)?
-        .with_intra_threads(num_cpus::get_physical())?
-        .with_parallel_execution(true)?
-        .with_inter_threads(num_cpus::get_physical())?
-        .commit_from_memory(model_file.as_ref())?)
-}
-
-#[allow(clippy::too_many_arguments)]
-pub fn synthesize(
-    session: &Session,
-    bert_ori: Array2<f32>,
-    x_tst: Array1<i64>,
-    sid: Array1<i64>,
-    tones: Array1<i64>,
-    lang_ids: Array1<i64>,
-    style_vector: Array1<f32>,
-    sdp_ratio: f32,
-    length_scale: f32,
-) -> Result<Array3<f32>> {
-    let bert = bert_ori.insert_axis(Axis(0));
-    let x_tst_lengths: Array1<i64> = array![x_tst.shape()[0] as i64];
-    let x_tst = x_tst.insert_axis(Axis(0));
-    let lang_ids = lang_ids.insert_axis(Axis(0));
-    let tones = tones.insert_axis(Axis(0));
-    let style_vector = style_vector.insert_axis(Axis(0));
-    let outputs = session.run(ort::inputs! {
-        "x_tst" => x_tst,
-        "x_tst_lengths" => x_tst_lengths,
-        "sid" => sid,
-        "tones" => tones,
-        "language" => lang_ids,
-        "bert" => bert,
-        "style_vec" => style_vector,
-        "sdp_ratio" => array![sdp_ratio],
-        "length_scale" => array![length_scale],
-    }?)?;
-
-    let audio_array = outputs["output"]
-        .try_extract_tensor::<f32>()?
-        .into_dimensionality::<Ix3>()?
-        .to_owned();
-
-    Ok(audio_array)
-}

sbv2_wasm/build.sh

@@ -1,4 +0,0 @@
-wasm-pack build --target web sbv2_wasm
-wasm-opt -O3 -o sbv2_wasm/pkg/sbv2_wasm_bg.wasm sbv2_wasm/pkg/sbv2_wasm_bg.wasm
-mkdir -p sbv2_wasm/dist
-cp sbv2_wasm/sbv2_wasm/pkg/sbv2_wasm_bg.wasm sbv2_wasm/dist/sbv2_wasm_bg.wasm

sbv2_wasm/example.html

@@ -1,51 +0,0 @@
-<!doctype html>
-<html lang="en">
-
-<head>
-    <meta charset="UTF-8" />
-    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
-    <title>Style Bert VITS2 Web</title>
-    <script type="importmap">
-        {
-            "imports": {
-            "onnxruntime-web": "https://cdn.jsdelivr.net/npm/onnxruntime-web@1.19.2/dist/ort.all.min.mjs",
-            "sbv2": "https://cdn.jsdelivr.net/npm/sbv2@0.1.1+esm"
-            }
-        }
-    </script>
-    <script type="module" async defer>
-        import { ModelHolder } from "sbv2";
-        await ModelHolder.globalInit(
-            await (
-                await fetch("https://esm.sh/sbv2@0.1.1/dist/sbv2_wasm_bg.wasm", { cache: "force-cache" })
-            ).arrayBuffer(),
-        );
-        const holder = await ModelHolder.create(
-            await (
-                await fetch("/models/tokenizer.json", { cache: "force-cache" })
-            ).text(),
-            await (
-                await fetch("/models/deberta.onnx", { cache: "force-cache" })
-            ).arrayBuffer(),
-        );
-        if (typeof window.onready == "function") {
-            window.onready(holder);
-        }
-    </script>
-    <script type="module" async defer>
-        window.onready = async function (holder) {
-            await holder.load(
-                "amitaro",
-                await (await fetch("/models/amitaro.sbv2")).arrayBuffer(),
-            );
-            const wave = await holder.synthesize("amitaro", "おはよう");
-            console.log(wave);
-        };
-    </script>
-</head>
-
-<body>
-    <div id="root"></div>
-</body>
-
-</html>

scripts/.gitignore

@@ -0,0 +1,5 @@
+*.json
+venv/
+tmp/
+*.safetensors
+*.npy

scripts/convert/convert_deberta.py

@@ -5,6 +5,7 @@ from transformers import AutoModelForMaskedLM, AutoTokenizer
 import torch
 from torch import nn
 from argparse import ArgumentParser
+import os
 
 parser = ArgumentParser()
 parser.add_argument("--model", default="ku-nlp/deberta-v2-large-japanese-char-wwm")
@@ -15,7 +16,7 @@ bert_models.load_tokenizer(Languages.JP, model_name)
 tokenizer = bert_models.load_tokenizer(Languages.JP)
 converter = BertConverter(tokenizer)
 tokenizer = converter.converted()
-tokenizer.save("../models/tokenizer.json")
+tokenizer.save("../../models/tokenizer.json")
 
 
 class ORTDeberta(nn.Module):
@@ -42,9 +43,10 @@ inputs = AutoTokenizer.from_pretrained(model_name)(
 torch.onnx.export(
     model,
     (inputs["input_ids"], inputs["token_type_ids"], inputs["attention_mask"]),
-    "../models/deberta.onnx",
+    "../../models/deberta.onnx",
     input_names=["input_ids", "token_type_ids", "attention_mask"],
     output_names=["output"],
     verbose=True,
     dynamic_axes={"input_ids": {1: "batch_size"}, "attention_mask": {1: "batch_size"}},
 )
+os.system("onnxsim ../../models/deberta.onnx ../../models/deberta.onnx")

scripts/convert/convert_model.py

@@ -36,7 +36,7 @@ data = array.tolist()
 hyper_parameters = HyperParameters.load_from_json(config_file)
 out_name = hyper_parameters.model_name
 
-with open(f"../models/style_vectors_{out_name}.json", "w") as f:
+with open(f"../../models/style_vectors_{out_name}.json", "w") as f:
     json.dump(
         {
             "data": data,
@@ -94,7 +94,7 @@ model = get_net_g(
 )
 
 
-def forward(x, x_len, sid, tone, lang, bert, style, length_scale, sdp_ratio):
+def forward(x, x_len, sid, tone, lang, bert, style, length_scale, sdp_ratio, noise_scale, noise_scale_w):
     return model.infer(
         x,
         x_len,
@@ -105,6 +105,8 @@ def forward(x, x_len, sid, tone, lang, bert, style, length_scale, sdp_ratio):
         style,
         sdp_ratio=sdp_ratio,
         length_scale=length_scale,
+        noise_scale=noise_scale,
+        noise_scale_w=noise_scale_w,
     )
 
 
@@ -122,8 +124,10 @@ torch.onnx.export(
         style_vec_tensor,
         torch.tensor(1.0),
         torch.tensor(0.0),
+        torch.tensor(0.6777),
+        torch.tensor(0.8),
     ),
-    f"../models/model_{out_name}.onnx",
+    f"../../models/model_{out_name}.onnx",
     verbose=True,
     dynamic_axes={
         "x_tst": {0: "batch_size", 1: "x_tst_max_length"},
@@ -144,14 +148,16 @@ torch.onnx.export(
         "style_vec",
         "length_scale",
         "sdp_ratio",
+        "noise_scale",
+        "noise_scale_w"
     ],
     output_names=["output"],
 )
-os.system(f"onnxsim ../models/model_{out_name}.onnx ../models/model_{out_name}.onnx")
-onnxfile = open(f"../models/model_{out_name}.onnx", "rb").read()
-stylefile = open(f"../models/style_vectors_{out_name}.json", "rb").read()
+os.system(f"onnxsim ../../models/model_{out_name}.onnx ../../models/model_{out_name}.onnx")
+onnxfile = open(f"../../models/model_{out_name}.onnx", "rb").read()
+stylefile = open(f"../../models/style_vectors_{out_name}.json", "rb").read()
 version = bytes("1", "utf8")
-with taropen(f"../models/tmp_{out_name}.sbv2tar", "w") as w:
+with taropen(f"../../models/tmp_{out_name}.sbv2tar", "w") as w:
 
     def add_tar(f, b):
         t = TarInfo(f)
@@ -161,9 +167,9 @@ with taropen(f"../models/tmp_{out_name}.sbv2tar", "w") as w:
     add_tar("version.txt", version)
     add_tar("model.onnx", onnxfile)
     add_tar("style_vectors.json", stylefile)
-open(f"../models/{out_name}.sbv2", "wb").write(
+open(f"../../models/{out_name}.sbv2", "wb").write(
     ZstdCompressor(threads=-1, level=22).compress(
-        open(f"../models/tmp_{out_name}.sbv2tar", "rb").read()
+        open(f"../../models/tmp_{out_name}.sbv2tar", "rb").read()
     )
 )
-os.unlink(f"../models/tmp_{out_name}.sbv2tar")
+os.unlink(f"../../models/tmp_{out_name}.sbv2tar")

scripts/make_dict.sh

@@ -0,0 +1,14 @@
+#!/bin/bash
+set -e
+git clone https://github.com/Aivis-Project/AivisSpeech-Engine ./scripts/tmp --filter=blob:none -n
+cd ./scripts/tmp
+git checkout 168b2a1144afe300b0490d9a6dd773ec6e927667 -- resources/dictionaries/*.csv
+cd ../..
+rm -rf ./crates/sbv2_core/src/dic
+cp -r ./scripts/tmp/resources/dictionaries ./crates/sbv2_core/src/dic
+rm -rf ./scripts/tmp
+for file in ./crates/sbv2_core/src/dic/0*.csv; do
+    /usr/bin/cat "$file"
+    echo
+done > ./crates/sbv2_core/src/all.csv
+lindera build ./crates/sbv2_core/src/all.csv ./crates/sbv2_core/src/dic/all.dic -u -k ipadic

scripts/sbv2-bindings-colab.ipynb

@@ -16,7 +16,7 @@
    "outputs": [],
    "source": [
     "# 必要なパッケージのインストール\n",
-    "!pip install sbv2_bindings\n",
+    "%pip install sbv2_bindings\n",
     "\n",
     "# 必要なモジュールのインポート\n",
     "import os\n",

scripts/sbv2-test-bindings.py

@@ -3,16 +3,16 @@ from sbv2_bindings import TTSModel
 
 def main():
     print("Loading models...")
-    model = TTSModel.from_path("../models/debert.onnx", "../models/tokenizer.json")
+    model = TTSModel.from_path("./models/debert.onnx", "./models/tokenizer.json")
     print("Models loaded!")
 
-    model.load_sbv2file_from_path("amitaro", "../models/amitaro.sbv2")
+    model.load_sbv2file_from_path("amitaro", "./models/amitaro.sbv2")
     print("All setup is done!")
 
     style_vector = model.get_style_vector("amitaro", 0, 1.0)
     with open("output.wav", "wb") as f:
         f.write(
-            model.synthesize("おはようございます。", "amitaro", style_vector, 0.0, 0.5)
+            model.synthesize("おはようございます。", "amitaro", 0, 0, 0.0, 0.5)
         )