From 7d0aa000509bba4df254fbc1d0a16286d9410f0d Mon Sep 17 00:00:00 2001
From: Ryo Yamashita <qryxip@gmail.com>
Date: Mon, 22 Jan 2024 02:59:31 +0900
Subject: [PATCH 1/7] =?UTF-8?q?`metas`=E5=87=BA=E5=8A=9B=E6=99=82=E3=81=AB?=
=?UTF-8?q?=E8=A9=B1=E8=80=85=E6=83=85=E5=A0=B1=E3=82=92=E3=83=9E=E3=83=BC?=
=?UTF-8?q?=E3=82=B8=E3=81=99=E3=82=8B?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../voicevox_core/src/__internal/interop.rs | 2 +-
crates/voicevox_core/src/infer/status.rs | 44 ++++++++++++-------
crates/voicevox_core/src/metas.rs | 41 +++++++++++++++++
.../src/compatible_engine.rs | 4 +-
4 files changed, 72 insertions(+), 19 deletions(-)
diff --git a/crates/voicevox_core/src/__internal/interop.rs b/crates/voicevox_core/src/__internal/interop.rs
index c8cd7101f..fe46d10bc 100644
--- a/crates/voicevox_core/src/__internal/interop.rs
+++ b/crates/voicevox_core/src/__internal/interop.rs
@@ -1 +1 @@
-pub use crate::synthesizer::blocking::PerformInference;
+pub use crate::{metas::merge as merge_metas, synthesizer::blocking::PerformInference};
diff --git a/crates/voicevox_core/src/infer/status.rs b/crates/voicevox_core/src/infer/status.rs
index 12367dda2..a9a1836e0 100644
--- a/crates/voicevox_core/src/infer/status.rs
+++ b/crates/voicevox_core/src/infer/status.rs
@@ -14,7 +14,7 @@ use crate::{
error::{ErrorRepr, LoadModelError, LoadModelErrorKind, LoadModelResult},
infer::{InferenceOperation, ParamInfo},
manifest::ModelInnerId,
- metas::{SpeakerMeta, StyleId, StyleMeta, VoiceModelMeta},
+ metas::{self, SpeakerMeta, StyleId, StyleMeta, VoiceModelMeta},
voice_model::{VoiceModelHeader, VoiceModelId},
Result,
};
@@ -130,11 +130,7 @@ struct LoadedModel<R: InferenceRuntime, D: InferenceDomain> {
impl<R: InferenceRuntime, D: InferenceDomain> LoadedModels<R, D> {
fn metas(&self) -> VoiceModelMeta {
- self.0
- .values()
- .flat_map(|LoadedModel { metas, .. }| metas)
- .cloned()
- .collect()
+ metas::merge(self.0.values().flat_map(|LoadedModel { metas, .. }| metas))
}
fn ids_for(&self, style_id: StyleId) -> Result<(VoiceModelId, ModelInnerId)> {
@@ -184,20 +180,33 @@ impl<R: InferenceRuntime, D: InferenceDomain> LoadedModels<R, D> {
///
/// # Errors
///
- /// 音声モデルIDかスタイルIDが`model_header`と重複するとき、エラーを返す。
+ /// 次の場合にエラーを返す。
+ ///
+ /// - 音声モデルIDかスタイルIDが`model_header`と重複するとき
+ /// - 同じ`speaker_uuid`で、スタイル以外でプロパティが異なっている話者があるとき
fn ensure_acceptable(&self, model_header: &VoiceModelHeader) -> LoadModelResult<()> {
- let loaded = self.styles();
- let external = model_header
- .metas
- .iter()
- .flat_map(|speaker| speaker.styles());
-
let error = |context| LoadModelError {
path: model_header.path.clone(),
context,
source: None,
};
+ let loaded = self.speakers();
+ let external = model_header.metas.iter();
+ if let Some((_loaded, _external)) =
+ iproduct!(loaded, external).find(|(loaded, external)| {
+ loaded.speaker_uuid() == external.speaker_uuid()
+ && !loaded.eq_except_styles(external)
+ })
+ {
+ todo!("same `speaker_uuid` but different properties");
+ }
+
+ let loaded = self.styles();
+ let external = model_header
+ .metas
+ .iter()
+ .flat_map(|speaker| speaker.styles());
if self.0.contains_key(&model_header.id) {
return Err(error(LoadModelErrorKind::ModelAlreadyLoaded {
id: model_header.id.clone(),
@@ -242,11 +251,12 @@ impl<R: InferenceRuntime, D: InferenceDomain> LoadedModels<R, D> {
Ok(())
}
+ fn speakers(&self) -> impl Iterator<Item = &SpeakerMeta> {
+ self.0.values().flat_map(|LoadedModel { metas, .. }| metas)
+ }
+
fn styles(&self) -> impl Iterator<Item = &StyleMeta> {
- self.0
- .values()
- .flat_map(|LoadedModel { metas, .. }| metas)
- .flat_map(|speaker| speaker.styles())
+ self.speakers().flat_map(|speaker| speaker.styles())
}
}
diff --git a/crates/voicevox_core/src/metas.rs b/crates/voicevox_core/src/metas.rs
index 77cb3a9fc..6e5ab29ea 100644
--- a/crates/voicevox_core/src/metas.rs
+++ b/crates/voicevox_core/src/metas.rs
@@ -2,8 +2,29 @@ use std::fmt::Display;
use derive_getters::Getters;
use derive_new::new;
+use itertools::Itertools as _;
use serde::{Deserialize, Serialize};
+pub fn merge<'a>(metas: impl IntoIterator<Item = &'a SpeakerMeta>) -> Vec<SpeakerMeta> {
+ metas
+ .into_iter()
+ .into_grouping_map_by(|speaker| &speaker.speaker_uuid)
+ .aggregate::<_, SpeakerMeta>(|acc, _, speaker| {
+ Some(
+ acc.map(|mut acc| {
+ acc.styles.extend(speaker.styles.clone());
+ acc
+ })
+ .unwrap_or_else(|| speaker.clone()),
+ )
+ })
+ .into_values()
+ .sorted_by_key(|SpeakerMeta { styles, .. }| {
+ styles.iter().map(|&StyleMeta { id, .. }| id).min()
+ })
+ .collect()
+}
+
/// [`StyleId`]の実体。
///
/// [`StyleId`]: StyleId
@@ -67,6 +88,26 @@ pub struct SpeakerMeta {
speaker_uuid: String,
}
+impl SpeakerMeta {
+ pub(crate) fn eq_except_styles(&self, other: &Self) -> bool {
+ let Self {
+ name: name1,
+ styles: _,
+ version: version1,
+ speaker_uuid: speaker_uuid1,
+ } = self;
+
+ let Self {
+ name: name2,
+ styles: _,
+ version: version2,
+ speaker_uuid: speaker_uuid2,
+ } = other;
+
+ (name1, version1, speaker_uuid1) == (name2, version2, speaker_uuid2)
+ }
+}
+
/// **スタイル**(_style_)のメタ情報。
#[derive(Deserialize, Serialize, Getters, Clone)]
pub struct StyleMeta {
diff --git a/crates/voicevox_core_c_api/src/compatible_engine.rs b/crates/voicevox_core_c_api/src/compatible_engine.rs
index 5a6f20f76..6755910f5 100644
--- a/crates/voicevox_core_c_api/src/compatible_engine.rs
+++ b/crates/voicevox_core_c_api/src/compatible_engine.rs
@@ -41,7 +41,9 @@ static VOICE_MODEL_SET: Lazy<VoiceModelSet> = Lazy::new(|| {
.iter()
.map(|vvm| (vvm.id().clone(), vvm.clone()))
.collect();
- let metas: Vec<_> = all_vvms.iter().flat_map(|vvm| vvm.metas()).collect();
+ let metas = voicevox_core::__internal::interop::merge_metas(
+ all_vvms.iter().flat_map(|vvm| vvm.metas()),
+ );
let mut style_model_map = BTreeMap::default();
for vvm in all_vvms.iter() {
for meta in vvm.metas().iter() {
From 2b417633abc16d8740f4dddab9b37f6703aa5236 Mon Sep 17 00:00:00 2001
From: Ryo Yamashita <qryxip@gmail.com>
Date: Mon, 22 Jan 2024 03:10:58 +0900
Subject: [PATCH 2/7] =?UTF-8?q?=E3=82=B9=E3=82=BF=E3=82=A4=E3=83=AB?=
=?UTF-8?q?=E3=82=82=E3=82=BD=E3=83=BC=E3=83=88=E3=81=99=E3=82=8B?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
crates/voicevox_core/src/metas.rs | 10 ++++++++--
1 file changed, 8 insertions(+), 2 deletions(-)
diff --git a/crates/voicevox_core/src/metas.rs b/crates/voicevox_core/src/metas.rs
index 6e5ab29ea..7dd943de7 100644
--- a/crates/voicevox_core/src/metas.rs
+++ b/crates/voicevox_core/src/metas.rs
@@ -19,8 +19,14 @@ pub fn merge<'a>(metas: impl IntoIterator<Item = &'a SpeakerMeta>) -> Vec<Speake
)
})
.into_values()
- .sorted_by_key(|SpeakerMeta { styles, .. }| {
- styles.iter().map(|&StyleMeta { id, .. }| id).min()
+ .map(|mut speaker| {
+ speaker
+ .styles
+ .sort_unstable_by_key(|&StyleMeta { id, .. }| id);
+ speaker
+ })
+ .sorted_unstable_by_key(|SpeakerMeta { styles, .. }| {
+ styles.first().map(|&StyleMeta { id, .. }| id)
})
.collect()
}
From 178d5ac29c8007100f7bfdb54f67c0b46a70bcc2 Mon Sep 17 00:00:00 2001
From: Ryo Yamashita <qryxip@gmail.com>
Date: Sat, 27 Jan 2024 16:30:52 +0900
Subject: [PATCH 3/7] =?UTF-8?q?`SpeakerMeta::{speaker=5Forder,style=5Forde?=
=?UTF-8?q?r}`=E3=82=92=E5=B0=8E=E5=85=A5?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
crates/voicevox_core/src/infer/status.rs | 3 +-
crates/voicevox_core/src/metas.rs | 61 ++++++++++++-------
crates/voicevox_core/src/voice_model.rs | 12 +++-
.../tests/e2e/snapshots.toml | 24 ++++++--
.../jp/hiroshiba/voicevoxcore/VoiceModel.java | 24 ++++++++
.../python/voicevox_core/_models.py | 16 ++++-
6 files changed, 110 insertions(+), 30 deletions(-)
diff --git a/crates/voicevox_core/src/infer/status.rs b/crates/voicevox_core/src/infer/status.rs
index a9a1836e0..bef8889d3 100644
--- a/crates/voicevox_core/src/infer/status.rs
+++ b/crates/voicevox_core/src/infer/status.rs
@@ -8,6 +8,7 @@ use std::{
use anyhow::bail;
use educe::Educe;
use enum_map::{Enum as _, EnumMap};
+use indexmap::IndexMap;
use itertools::{iproduct, Itertools as _};
use crate::{
@@ -119,7 +120,7 @@ impl<R: InferenceRuntime, D: InferenceDomain> Status<R, D> {
#[derive(Educe)]
#[educe(Default(bound = "R: InferenceRuntime, D: InferenceDomain"))]
struct LoadedModels<R: InferenceRuntime, D: InferenceDomain>(
- BTreeMap<VoiceModelId, LoadedModel<R, D>>,
+ IndexMap<VoiceModelId, LoadedModel<R, D>>,
);
struct LoadedModel<R: InferenceRuntime, D: InferenceDomain> {
diff --git a/crates/voicevox_core/src/metas.rs b/crates/voicevox_core/src/metas.rs
index 7dd943de7..47682c2ad 100644
--- a/crates/voicevox_core/src/metas.rs
+++ b/crates/voicevox_core/src/metas.rs
@@ -2,33 +2,36 @@ use std::fmt::Display;
use derive_getters::Getters;
use derive_new::new;
+use indexmap::{IndexMap, IndexSet};
use itertools::Itertools as _;
use serde::{Deserialize, Serialize};
pub fn merge<'a>(metas: impl IntoIterator<Item = &'a SpeakerMeta>) -> Vec<SpeakerMeta> {
- metas
+ return metas
.into_iter()
- .into_grouping_map_by(|speaker| &speaker.speaker_uuid)
- .aggregate::<_, SpeakerMeta>(|acc, _, speaker| {
- Some(
- acc.map(|mut acc| {
- acc.styles.extend(speaker.styles.clone());
- acc
- })
- .unwrap_or_else(|| speaker.clone()),
- )
+ .fold(IndexMap::<_, SpeakerMeta>::new(), |mut acc, speaker| {
+ acc.entry(&speaker.speaker_uuid)
+ .and_modify(|acc| acc.styles.extend(speaker.styles.clone()))
+ .or_insert_with(|| speaker.clone());
+ acc
})
.into_values()
- .map(|mut speaker| {
- speaker
- .styles
- .sort_unstable_by_key(|&StyleMeta { id, .. }| id);
- speaker
+ .update(|speaker| {
+ speaker.styles.sort_by_key(|StyleMeta { id, .. }| {
+ key(speaker
+ .style_order
+ .get_index_of(id)
+ .map(|i| i.try_into().unwrap()))
+ });
})
- .sorted_unstable_by_key(|SpeakerMeta { styles, .. }| {
- styles.first().map(|&StyleMeta { id, .. }| id)
- })
- .collect()
+ .sorted_by_key(|&SpeakerMeta { speaker_order, .. }| key(speaker_order))
+ .collect();
+
+ fn key(order: Option<u32>) -> impl Ord {
+ order
+ .map(Into::into)
+ .unwrap_or_else(|| u64::from(u32::MAX) + 1)
+ }
}
/// [`StyleId`]の実体。
@@ -42,7 +45,7 @@ pub type RawStyleId = u32;
///
/// [**話者**(_speaker_)]: SpeakerMeta
/// [**スタイル**(_style_)]: StyleMeta
-#[derive(PartialEq, Eq, Clone, Copy, Ord, PartialOrd, Deserialize, Serialize, new, Debug)]
+#[derive(PartialEq, Eq, Clone, Copy, Ord, Hash, PartialOrd, Deserialize, Serialize, new, Debug)]
pub struct StyleId(RawStyleId);
impl StyleId {
@@ -92,6 +95,17 @@ pub struct SpeakerMeta {
version: StyleVersion,
/// 話者のUUID。
speaker_uuid: String,
+ /// 話者の順番。
+ ///
+ /// `SpeakerMeta`の列は、この値に対して昇順に並んでいるべきである。
+ speaker_order: Option<u32>,
+ /// 話者に属するスタイルの順番。
+ ///
+ /// [`styles`]はこの並びに沿うべきである。
+ ///
+ /// [`styles`]: Self::styles
+ #[serde(default)]
+ style_order: IndexSet<StyleId>,
}
impl SpeakerMeta {
@@ -101,6 +115,8 @@ impl SpeakerMeta {
styles: _,
version: version1,
speaker_uuid: speaker_uuid1,
+ speaker_order: speaker_order1,
+ style_order: style_order1,
} = self;
let Self {
@@ -108,9 +124,12 @@ impl SpeakerMeta {
styles: _,
version: version2,
speaker_uuid: speaker_uuid2,
+ speaker_order: speaker_order2,
+ style_order: style_order2,
} = other;
- (name1, version1, speaker_uuid1) == (name2, version2, speaker_uuid2)
+ (name1, version1, speaker_uuid1, speaker_order1, style_order1)
+ == (name2, version2, speaker_uuid2, speaker_order2, style_order2)
}
}
diff --git a/crates/voicevox_core/src/voice_model.rs b/crates/voicevox_core/src/voice_model.rs
index 96bf481d1..fc8f4d20f 100644
--- a/crates/voicevox_core/src/voice_model.rs
+++ b/crates/voicevox_core/src/voice_model.rs
@@ -15,7 +15,17 @@ pub type RawVoiceModelId = String;
/// 音声モデルID。
#[derive(
- PartialEq, Eq, Clone, Ord, PartialOrd, Deserialize, new, Getters, derive_more::Display, Debug,
+ PartialEq,
+ Eq,
+ Clone,
+ Ord,
+ Hash,
+ PartialOrd,
+ Deserialize,
+ new,
+ Getters,
+ derive_more::Display,
+ Debug,
)]
pub struct VoiceModelId {
raw_voice_model_id: RawVoiceModelId,
diff --git a/crates/voicevox_core_c_api/tests/e2e/snapshots.toml b/crates/voicevox_core_c_api/tests/e2e/snapshots.toml
index 8f3fa4f3b..d0e80f97f 100644
--- a/crates/voicevox_core_c_api/tests/e2e/snapshots.toml
+++ b/crates/voicevox_core_c_api/tests/e2e/snapshots.toml
@@ -10,7 +10,9 @@ metas = '''
}
],
"version": "0.0.1",
- "speaker_uuid": "574bc678-8370-44be-b941-08e46e7b47d7"
+ "speaker_uuid": "574bc678-8370-44be-b941-08e46e7b47d7",
+ "speaker_order": null,
+ "style_order": []
},
{
"name": "dummy2",
@@ -21,7 +23,9 @@ metas = '''
}
],
"version": "0.0.1",
- "speaker_uuid": "dd9ccd75-75f6-40ce-a3db-960cbed2e905"
+ "speaker_uuid": "dd9ccd75-75f6-40ce-a3db-960cbed2e905",
+ "speaker_order": null,
+ "style_order": []
},
{
"name": "dummy3",
@@ -36,7 +40,9 @@ metas = '''
}
],
"version": "0.0.1",
- "speaker_uuid": "5d3d9aa9-88e5-4a96-8ef7-f13a3cad1cb3"
+ "speaker_uuid": "5d3d9aa9-88e5-4a96-8ef7-f13a3cad1cb3",
+ "speaker_order": null,
+ "style_order": []
}
]'''
stderr.windows = '''
@@ -93,7 +99,9 @@ metas = '''
}
],
"version": "0.0.1",
- "speaker_uuid": "574bc678-8370-44be-b941-08e46e7b47d7"
+ "speaker_uuid": "574bc678-8370-44be-b941-08e46e7b47d7",
+ "speaker_order": null,
+ "style_order": []
},
{
"name": "dummy2",
@@ -104,7 +112,9 @@ metas = '''
}
],
"version": "0.0.1",
- "speaker_uuid": "dd9ccd75-75f6-40ce-a3db-960cbed2e905"
+ "speaker_uuid": "dd9ccd75-75f6-40ce-a3db-960cbed2e905",
+ "speaker_order": null,
+ "style_order": []
},
{
"name": "dummy3",
@@ -119,7 +129,9 @@ metas = '''
}
],
"version": "0.0.1",
- "speaker_uuid": "5d3d9aa9-88e5-4a96-8ef7-f13a3cad1cb3"
+ "speaker_uuid": "5d3d9aa9-88e5-4a96-8ef7-f13a3cad1cb3",
+ "speaker_order": null,
+ "style_order": []
}
]'''
stderr.windows = '''
diff --git a/crates/voicevox_core_java_api/lib/src/main/java/jp/hiroshiba/voicevoxcore/VoiceModel.java b/crates/voicevox_core_java_api/lib/src/main/java/jp/hiroshiba/voicevoxcore/VoiceModel.java
index 05c1a11b2..0e5715e09 100644
--- a/crates/voicevox_core_java_api/lib/src/main/java/jp/hiroshiba/voicevoxcore/VoiceModel.java
+++ b/crates/voicevox_core_java_api/lib/src/main/java/jp/hiroshiba/voicevoxcore/VoiceModel.java
@@ -4,6 +4,8 @@
import com.google.gson.annotations.Expose;
import com.google.gson.annotations.SerializedName;
import jakarta.annotation.Nonnull;
+import jakarta.annotation.Nullable;
+import java.util.LinkedHashSet;
/** 音声モデル。 */
public class VoiceModel extends Dll {
@@ -68,6 +70,26 @@ public static class SpeakerMeta {
@Nonnull
public final String version;
+ /**
+ * 話者の順番。
+ *
+ * <p>{@code SpeakerMeta}の列は、この値に対して昇順に並んでいるべきである。
+ */
+ @SerializedName("speaker_order")
+ @Expose
+ @Nullable
+ public final Integer speakerOrder;
+
+ /**
+ * 話者に属するスタイルの順番。
+ *
+ * <p>{@link #styles}はこの並びに沿うべきである。
+ */
+ @SerializedName("style_order")
+ @Expose
+ @Nonnull
+ public final LinkedHashSet<Integer> styleOrder;
+
private SpeakerMeta() {
// GSONからコンストラクトするため、このメソッドは呼ばれることは無い。
// このメソッドは@Nonnullを満たすために必要。
@@ -75,6 +97,8 @@ private SpeakerMeta() {
this.styles = new StyleMeta[0];
this.speakerUuid = "";
this.version = "";
+ this.speakerOrder = null;
+ this.styleOrder = new LinkedHashSet<>();
}
}
diff --git a/crates/voicevox_core_python_api/python/voicevox_core/_models.py b/crates/voicevox_core_python_api/python/voicevox_core/_models.py
index 195154629..9e11e8bd8 100644
--- a/crates/voicevox_core_python_api/python/voicevox_core/_models.py
+++ b/crates/voicevox_core_python_api/python/voicevox_core/_models.py
@@ -1,6 +1,6 @@
import dataclasses
from enum import Enum
-from typing import List, NewType, Optional
+from typing import Dict, List, NewType, Optional
import pydantic
@@ -61,6 +61,20 @@ class SpeakerMeta:
version: StyleVersion
"""話者のUUID。"""
+ speaker_order: Optional[int] = None
+ """
+ 話者の順番。
+
+ ``SpeakerMeta`` の列は、この値に対して昇順に並んでいるべきである。
+ """
+
+ style_order: List[StyleId] = dataclasses.field(default_factory=list)
+ """
+ 話者に属するスタイルの順番。
+
+ :attr:`styles` はこの並びに沿うべきである。
+ """
+
@pydantic.dataclasses.dataclass
class SupportedDevices:
From e78dd4f5228664b1ccaf67d7ceb264d3ff0164d3 Mon Sep 17 00:00:00 2001
From: Ryo Yamashita <qryxip@gmail.com>
Date: Sun, 28 Jan 2024 00:31:52 +0900
Subject: [PATCH 4/7] =?UTF-8?q?=E3=83=86=E3=82=B9=E3=83=88=E3=82=92?=
=?UTF-8?q?=E8=BF=BD=E5=8A=A0?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
crates/voicevox_core/src/metas.rs | 118 ++++++++++++++++++++++++++++++
1 file changed, 118 insertions(+)
diff --git a/crates/voicevox_core/src/metas.rs b/crates/voicevox_core/src/metas.rs
index 47682c2ad..22426f5fe 100644
--- a/crates/voicevox_core/src/metas.rs
+++ b/crates/voicevox_core/src/metas.rs
@@ -141,3 +141,121 @@ pub struct StyleMeta {
/// スタイル名。
name: String,
}
+
+#[cfg(test)]
+mod tests {
+ use once_cell::sync::Lazy;
+ use serde_json::json;
+
+ #[test]
+ fn merge_works() -> anyhow::Result<()> {
+ static INPUT: Lazy<serde_json::Value> = Lazy::new(|| {
+ json!([
+ {
+ "name": "B",
+ "styles": [
+ {
+ "id": 3,
+ "name": "B_1"
+ }
+ ],
+ "version": "0.0.0",
+ "speaker_uuid": "f34ab151-c0f5-4e0a-9ad2-51ce30dba24d",
+ "speaker_order": 1,
+ "style_order": [
+ 3
+ ]
+ },
+ {
+ "name": "A",
+ "styles": [
+ {
+ "id": 2,
+ "name": "A_3"
+ }
+ ],
+ "version": "0.0.0",
+ "speaker_uuid": "d6fd707c-a451-48e9-8f00-fe9ee3bf6264",
+ "speaker_order": 0,
+ "style_order": [
+ 1,
+ 0,
+ 2
+ ]
+ },
+ {
+ "name": "A",
+ "styles": [
+ {
+ "id": 1,
+ "name": "A_1"
+ },
+ {
+ "id": 0,
+ "name": "A_2"
+ }
+ ],
+ "version": "0.0.0",
+ "speaker_uuid": "d6fd707c-a451-48e9-8f00-fe9ee3bf6264",
+ "speaker_order": 0,
+ "style_order": [
+ 1,
+ 0,
+ 2
+ ]
+ }
+ ])
+ });
+
+ static EXPECTED: Lazy<serde_json::Value> = Lazy::new(|| {
+ json!([
+ {
+ "name": "A",
+ "styles": [
+ {
+ "id": 1,
+ "name": "A_1"
+ },
+ {
+ "id": 0,
+ "name": "A_2"
+ },
+ {
+ "id": 2,
+ "name": "A_3"
+ }
+ ],
+ "version": "0.0.0",
+ "speaker_uuid": "d6fd707c-a451-48e9-8f00-fe9ee3bf6264",
+ "speaker_order": 0,
+ "style_order": [
+ 1,
+ 0,
+ 2
+ ]
+ },
+ {
+ "name": "B",
+ "styles": [
+ {
+ "id": 3,
+ "name": "B_1"
+ }
+ ],
+ "version": "0.0.0",
+ "speaker_uuid": "f34ab151-c0f5-4e0a-9ad2-51ce30dba24d",
+ "speaker_order": 1,
+ "style_order": [
+ 3
+ ]
+ }
+ ])
+ });
+
+ let input = &serde_json::from_value::<Vec<_>>(INPUT.clone())?;
+ let actual = serde_json::to_value(&super::merge(input))?;
+
+ pretty_assertions::assert_eq!(*EXPECTED, actual);
+ Ok(())
+ }
+}
From 9d2b1a2427f8dbccc397626d77befa2977bc7ab1 Mon Sep 17 00:00:00 2001
From: Ryo Yamashita <qryxip@gmail.com>
Date: Sun, 28 Jan 2024 00:42:33 +0900
Subject: [PATCH 5/7] =?UTF-8?q?`merge`=E3=81=ABdoc?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
crates/voicevox_core/src/metas.rs | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/crates/voicevox_core/src/metas.rs b/crates/voicevox_core/src/metas.rs
index 22426f5fe..8cd67c976 100644
--- a/crates/voicevox_core/src/metas.rs
+++ b/crates/voicevox_core/src/metas.rs
@@ -6,6 +6,15 @@ use indexmap::{IndexMap, IndexSet};
use itertools::Itertools as _;
use serde::{Deserialize, Serialize};
+/// [`speaker_uuid`]をキーとして複数の[`SpeakerMeta`]をマージする。
+///
+/// マージする際話者は[`speaker_order`]、スタイルは[`style_order`]をもとに安定ソートされる。
+/// `speaker_order`が無い話者と`style_order`に属さないスタイルは、そうでないものよりも後ろに
+/// 置かれる。
+///
+/// [`speaker_uuid`]: SpeakerMeta::speaker_uuid
+/// [`speaker_order`]: SpeakerMeta::speaker_order
+/// [`style_order`]: SpeakerMeta::style_order
pub fn merge<'a>(metas: impl IntoIterator<Item = &'a SpeakerMeta>) -> Vec<SpeakerMeta> {
return metas
.into_iter()
From 25a918cdbb6ddb3c01f4b0a511195c81e5ad4902 Mon Sep 17 00:00:00 2001
From: Ryo Yamashita <qryxip@gmail.com>
Date: Sun, 28 Jan 2024 00:56:31 +0900
Subject: [PATCH 6/7] =?UTF-8?q?=E8=A9=B1=E8=80=85=E6=83=85=E5=A0=B1?=
=?UTF-8?q?=E3=81=8C=E9=A3=9F=E3=81=84=E9=81=95=E3=81=86=E3=82=84=E3=81=A4?=
=?UTF-8?q?=E3=81=AFwarning=E6=AD=A2=E3=81=BE=E3=82=8A=E3=81=AB=E3=81=99?=
=?UTF-8?q?=E3=82=8B?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
crates/voicevox_core/src/infer/status.rs | 14 ++++------
crates/voicevox_core/src/metas.rs | 34 +++++++++++++++++++++---
2 files changed, 35 insertions(+), 13 deletions(-)
diff --git a/crates/voicevox_core/src/infer/status.rs b/crates/voicevox_core/src/infer/status.rs
index bef8889d3..db529e2e5 100644
--- a/crates/voicevox_core/src/infer/status.rs
+++ b/crates/voicevox_core/src/infer/status.rs
@@ -184,7 +184,6 @@ impl<R: InferenceRuntime, D: InferenceDomain> LoadedModels<R, D> {
/// 次の場合にエラーを返す。
///
/// - 音声モデルIDかスタイルIDが`model_header`と重複するとき
- /// - 同じ`speaker_uuid`で、スタイル以外でプロパティが異なっている話者があるとき
fn ensure_acceptable(&self, model_header: &VoiceModelHeader) -> LoadModelResult<()> {
let error = |context| LoadModelError {
path: model_header.path.clone(),
@@ -194,13 +193,10 @@ impl<R: InferenceRuntime, D: InferenceDomain> LoadedModels<R, D> {
let loaded = self.speakers();
let external = model_header.metas.iter();
- if let Some((_loaded, _external)) =
- iproduct!(loaded, external).find(|(loaded, external)| {
- loaded.speaker_uuid() == external.speaker_uuid()
- && !loaded.eq_except_styles(external)
- })
- {
- todo!("same `speaker_uuid` but different properties");
+ for (loaded, external) in iproduct!(loaded, external) {
+ if loaded.speaker_uuid() == external.speaker_uuid() {
+ loaded.warn_diff_except_styles(external);
+ }
}
let loaded = self.styles();
@@ -252,7 +248,7 @@ impl<R: InferenceRuntime, D: InferenceDomain> LoadedModels<R, D> {
Ok(())
}
- fn speakers(&self) -> impl Iterator<Item = &SpeakerMeta> {
+ fn speakers(&self) -> impl Iterator<Item = &SpeakerMeta> + Clone {
self.0.values().flat_map(|LoadedModel { metas, .. }| metas)
}
diff --git a/crates/voicevox_core/src/metas.rs b/crates/voicevox_core/src/metas.rs
index 8cd67c976..ae47e1803 100644
--- a/crates/voicevox_core/src/metas.rs
+++ b/crates/voicevox_core/src/metas.rs
@@ -1,10 +1,11 @@
-use std::fmt::Display;
+use std::fmt::{Debug, Display};
use derive_getters::Getters;
use derive_new::new;
use indexmap::{IndexMap, IndexSet};
use itertools::Itertools as _;
use serde::{Deserialize, Serialize};
+use tracing::warn;
/// [`speaker_uuid`]をキーとして複数の[`SpeakerMeta`]をマージする。
///
@@ -118,7 +119,10 @@ pub struct SpeakerMeta {
}
impl SpeakerMeta {
- pub(crate) fn eq_except_styles(&self, other: &Self) -> bool {
+ /// # Panics
+ ///
+ /// `speaker_uuid`が異なるときパニックする。
+ pub(crate) fn warn_diff_except_styles(&self, other: &Self) {
let Self {
name: name1,
styles: _,
@@ -137,8 +141,30 @@ impl SpeakerMeta {
style_order: style_order2,
} = other;
- (name1, version1, speaker_uuid1, speaker_order1, style_order1)
- == (name2, version2, speaker_uuid2, speaker_order2, style_order2)
+ if speaker_uuid1 != speaker_uuid2 {
+ panic!("must be equal: {speaker_uuid1} != {speaker_uuid2:?}");
+ }
+
+ warn_diff(speaker_uuid1, "name", name1, name2);
+ warn_diff(speaker_uuid1, "version", version1, version2);
+ warn_diff(
+ speaker_uuid1,
+ "speaker_order",
+ speaker_order1,
+ speaker_order2,
+ );
+ warn_diff(speaker_uuid1, "style_order", style_order1, style_order2);
+
+ fn warn_diff<T: PartialEq + Debug>(
+ speaker_uuid: &str,
+ field_name: &str,
+ left: &T,
+ right: &T,
+ ) {
+ if left != right {
+ warn!("`{speaker_uuid}`: different `{field_name}` ({left:?} != {right:?})");
+ }
+ }
}
}
From 40616781b590898efc7e9408c6c8dc189e751080 Mon Sep 17 00:00:00 2001
From: Ryo Yamashita <qryxip@gmail.com>
Date: Wed, 7 Feb 2024 02:10:49 +0900
Subject: [PATCH 7/7] =?UTF-8?q?`StyleMeta`=E8=87=AA=E4=BD=93=E3=81=AB`orde?=
=?UTF-8?q?r`=E3=82=92=E6=8C=81=E3=81=9F=E3=81=9B=E3=82=8B?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
crates/voicevox_core/src/metas.rs | 104 +++++++-----------
.../tests/e2e/snapshots.toml | 42 +++----
.../jp/hiroshiba/voicevoxcore/VoiceModel.java | 29 +++--
.../python/voicevox_core/_models.py | 18 +--
4 files changed, 82 insertions(+), 111 deletions(-)
diff --git a/crates/voicevox_core/src/metas.rs b/crates/voicevox_core/src/metas.rs
index ae47e1803..78314d52a 100644
--- a/crates/voicevox_core/src/metas.rs
+++ b/crates/voicevox_core/src/metas.rs
@@ -2,20 +2,17 @@ use std::fmt::{Debug, Display};
use derive_getters::Getters;
use derive_new::new;
-use indexmap::{IndexMap, IndexSet};
+use indexmap::IndexMap;
use itertools::Itertools as _;
use serde::{Deserialize, Serialize};
use tracing::warn;
/// [`speaker_uuid`]をキーとして複数の[`SpeakerMeta`]をマージする。
///
-/// マージする際話者は[`speaker_order`]、スタイルは[`style_order`]をもとに安定ソートされる。
-/// `speaker_order`が無い話者と`style_order`に属さないスタイルは、そうでないものよりも後ろに
-/// 置かれる。
+/// マージする際話者は[`SpeakerMeta::order`]、スタイルは[`StyleMeta::order`]をもとに安定ソートされる。
+/// `order`が無い話者とスタイルは、そうでないものよりも後ろに置かれる。
///
/// [`speaker_uuid`]: SpeakerMeta::speaker_uuid
-/// [`speaker_order`]: SpeakerMeta::speaker_order
-/// [`style_order`]: SpeakerMeta::style_order
pub fn merge<'a>(metas: impl IntoIterator<Item = &'a SpeakerMeta>) -> Vec<SpeakerMeta> {
return metas
.into_iter()
@@ -27,14 +24,11 @@ pub fn merge<'a>(metas: impl IntoIterator<Item = &'a SpeakerMeta>) -> Vec<Speake
})
.into_values()
.update(|speaker| {
- speaker.styles.sort_by_key(|StyleMeta { id, .. }| {
- key(speaker
- .style_order
- .get_index_of(id)
- .map(|i| i.try_into().unwrap()))
- });
+ speaker
+ .styles
+ .sort_by_key(|&StyleMeta { order, .. }| key(order));
})
- .sorted_by_key(|&SpeakerMeta { speaker_order, .. }| key(speaker_order))
+ .sorted_by_key(|&SpeakerMeta { order, .. }| key(order))
.collect();
fn key(order: Option<u32>) -> impl Ord {
@@ -108,14 +102,7 @@ pub struct SpeakerMeta {
/// 話者の順番。
///
/// `SpeakerMeta`の列は、この値に対して昇順に並んでいるべきである。
- speaker_order: Option<u32>,
- /// 話者に属するスタイルの順番。
- ///
- /// [`styles`]はこの並びに沿うべきである。
- ///
- /// [`styles`]: Self::styles
- #[serde(default)]
- style_order: IndexSet<StyleId>,
+ order: Option<u32>,
}
impl SpeakerMeta {
@@ -128,8 +115,7 @@ impl SpeakerMeta {
styles: _,
version: version1,
speaker_uuid: speaker_uuid1,
- speaker_order: speaker_order1,
- style_order: style_order1,
+ order: order1,
} = self;
let Self {
@@ -137,8 +123,7 @@ impl SpeakerMeta {
styles: _,
version: version2,
speaker_uuid: speaker_uuid2,
- speaker_order: speaker_order2,
- style_order: style_order2,
+ order: order2,
} = other;
if speaker_uuid1 != speaker_uuid2 {
@@ -147,13 +132,7 @@ impl SpeakerMeta {
warn_diff(speaker_uuid1, "name", name1, name2);
warn_diff(speaker_uuid1, "version", version1, version2);
- warn_diff(
- speaker_uuid1,
- "speaker_order",
- speaker_order1,
- speaker_order2,
- );
- warn_diff(speaker_uuid1, "style_order", style_order1, style_order2);
+ warn_diff(speaker_uuid1, "order", order1, order2);
fn warn_diff<T: PartialEq + Debug>(
speaker_uuid: &str,
@@ -175,6 +154,10 @@ pub struct StyleMeta {
id: StyleId,
/// スタイル名。
name: String,
+ /// スタイルの順番。
+ ///
+ /// [`SpeakerMeta::styles`]は、この値に対して昇順に並んでいるべきである。
+ order: Option<u32>,
}
#[cfg(test)]
@@ -191,53 +174,44 @@ mod tests {
"styles": [
{
"id": 3,
- "name": "B_1"
+ "name": "B_1",
+ "order": 0
}
],
"version": "0.0.0",
"speaker_uuid": "f34ab151-c0f5-4e0a-9ad2-51ce30dba24d",
- "speaker_order": 1,
- "style_order": [
- 3
- ]
+ "order": 1
},
{
"name": "A",
"styles": [
{
"id": 2,
- "name": "A_3"
+ "name": "A_3",
+ "order": 2
}
],
"version": "0.0.0",
"speaker_uuid": "d6fd707c-a451-48e9-8f00-fe9ee3bf6264",
- "speaker_order": 0,
- "style_order": [
- 1,
- 0,
- 2
- ]
+ "order": 0
},
{
"name": "A",
"styles": [
{
"id": 1,
- "name": "A_1"
+ "name": "A_1",
+ "order": 0
},
{
"id": 0,
- "name": "A_2"
+ "name": "A_2",
+ "order": 1
}
],
"version": "0.0.0",
"speaker_uuid": "d6fd707c-a451-48e9-8f00-fe9ee3bf6264",
- "speaker_order": 0,
- "style_order": [
- 1,
- 0,
- 2
- ]
+ "order": 0
}
])
});
@@ -249,46 +223,42 @@ mod tests {
"styles": [
{
"id": 1,
- "name": "A_1"
+ "name": "A_1",
+ "order": 0
},
{
"id": 0,
- "name": "A_2"
+ "name": "A_2",
+ "order": 1
},
{
"id": 2,
- "name": "A_3"
+ "name": "A_3",
+ "order": 2
}
],
"version": "0.0.0",
"speaker_uuid": "d6fd707c-a451-48e9-8f00-fe9ee3bf6264",
- "speaker_order": 0,
- "style_order": [
- 1,
- 0,
- 2
- ]
+ "order": 0
},
{
"name": "B",
"styles": [
{
"id": 3,
- "name": "B_1"
+ "name": "B_1",
+ "order": 0
}
],
"version": "0.0.0",
"speaker_uuid": "f34ab151-c0f5-4e0a-9ad2-51ce30dba24d",
- "speaker_order": 1,
- "style_order": [
- 3
- ]
+ "order": 1
}
])
});
let input = &serde_json::from_value::<Vec<_>>(INPUT.clone())?;
- let actual = serde_json::to_value(&super::merge(input))?;
+ let actual = serde_json::to_value(super::merge(input))?;
pretty_assertions::assert_eq!(*EXPECTED, actual);
Ok(())
diff --git a/crates/voicevox_core_c_api/tests/e2e/snapshots.toml b/crates/voicevox_core_c_api/tests/e2e/snapshots.toml
index d0e80f97f..25926487e 100644
--- a/crates/voicevox_core_c_api/tests/e2e/snapshots.toml
+++ b/crates/voicevox_core_c_api/tests/e2e/snapshots.toml
@@ -6,43 +6,44 @@ metas = '''
"styles": [
{
"id": 0,
- "name": "style1"
+ "name": "style1",
+ "order": null
}
],
"version": "0.0.1",
"speaker_uuid": "574bc678-8370-44be-b941-08e46e7b47d7",
- "speaker_order": null,
- "style_order": []
+ "order": null
},
{
"name": "dummy2",
"styles": [
{
"id": 1,
- "name": "style2"
+ "name": "style2",
+ "order": null
}
],
"version": "0.0.1",
"speaker_uuid": "dd9ccd75-75f6-40ce-a3db-960cbed2e905",
- "speaker_order": null,
- "style_order": []
+ "order": null
},
{
"name": "dummy3",
"styles": [
{
"id": 302,
- "name": "style3-1"
+ "name": "style3-1",
+ "order": null
},
{
"id": 303,
- "name": "style3-2"
+ "name": "style3-2",
+ "order": null
}
],
"version": "0.0.1",
"speaker_uuid": "5d3d9aa9-88e5-4a96-8ef7-f13a3cad1cb3",
- "speaker_order": null,
- "style_order": []
+ "order": null
}
]'''
stderr.windows = '''
@@ -95,43 +96,44 @@ metas = '''
"styles": [
{
"id": 0,
- "name": "style1"
+ "name": "style1",
+ "order": null
}
],
"version": "0.0.1",
"speaker_uuid": "574bc678-8370-44be-b941-08e46e7b47d7",
- "speaker_order": null,
- "style_order": []
+ "order": null
},
{
"name": "dummy2",
"styles": [
{
"id": 1,
- "name": "style2"
+ "name": "style2",
+ "order": null
}
],
"version": "0.0.1",
"speaker_uuid": "dd9ccd75-75f6-40ce-a3db-960cbed2e905",
- "speaker_order": null,
- "style_order": []
+ "order": null
},
{
"name": "dummy3",
"styles": [
{
"id": 302,
- "name": "style3-1"
+ "name": "style3-1",
+ "order": null
},
{
"id": 303,
- "name": "style3-2"
+ "name": "style3-2",
+ "order": null
}
],
"version": "0.0.1",
"speaker_uuid": "5d3d9aa9-88e5-4a96-8ef7-f13a3cad1cb3",
- "speaker_order": null,
- "style_order": []
+ "order": null
}
]'''
stderr.windows = '''
diff --git a/crates/voicevox_core_java_api/lib/src/main/java/jp/hiroshiba/voicevoxcore/VoiceModel.java b/crates/voicevox_core_java_api/lib/src/main/java/jp/hiroshiba/voicevoxcore/VoiceModel.java
index 0e5715e09..ba4881566 100644
--- a/crates/voicevox_core_java_api/lib/src/main/java/jp/hiroshiba/voicevoxcore/VoiceModel.java
+++ b/crates/voicevox_core_java_api/lib/src/main/java/jp/hiroshiba/voicevoxcore/VoiceModel.java
@@ -5,7 +5,6 @@
import com.google.gson.annotations.SerializedName;
import jakarta.annotation.Nonnull;
import jakarta.annotation.Nullable;
-import java.util.LinkedHashSet;
/** 音声モデル。 */
public class VoiceModel extends Dll {
@@ -75,20 +74,10 @@ public static class SpeakerMeta {
*
* <p>{@code SpeakerMeta}の列は、この値に対して昇順に並んでいるべきである。
*/
- @SerializedName("speaker_order")
+ @SerializedName("order")
@Expose
@Nullable
- public final Integer speakerOrder;
-
- /**
- * 話者に属するスタイルの順番。
- *
- * <p>{@link #styles}はこの並びに沿うべきである。
- */
- @SerializedName("style_order")
- @Expose
- @Nonnull
- public final LinkedHashSet<Integer> styleOrder;
+ public final Integer order;
private SpeakerMeta() {
// GSONからコンストラクトするため、このメソッドは呼ばれることは無い。
@@ -97,8 +86,7 @@ private SpeakerMeta() {
this.styles = new StyleMeta[0];
this.speakerUuid = "";
this.version = "";
- this.speakerOrder = null;
- this.styleOrder = new LinkedHashSet<>();
+ this.order = null;
}
}
@@ -115,9 +103,20 @@ public static class StyleMeta {
@Expose
public final int id;
+ /**
+ * 話者の順番。
+ *
+ * <p>{@link SpeakerMeta#styles}の列は、この値に対して昇順に並んでいるべきである。
+ */
+ @SerializedName("order")
+ @Expose
+ @Nullable
+ public final Integer order;
+
private StyleMeta() {
this.name = "";
this.id = 0;
+ this.order = null;
}
}
}
diff --git a/crates/voicevox_core_python_api/python/voicevox_core/_models.py b/crates/voicevox_core_python_api/python/voicevox_core/_models.py
index 9e11e8bd8..c72bbcbf0 100644
--- a/crates/voicevox_core_python_api/python/voicevox_core/_models.py
+++ b/crates/voicevox_core_python_api/python/voicevox_core/_models.py
@@ -1,6 +1,6 @@
import dataclasses
from enum import Enum
-from typing import Dict, List, NewType, Optional
+from typing import List, NewType, Optional
import pydantic
@@ -44,6 +44,13 @@ class StyleMeta:
id: StyleId
"""スタイルID。"""
+ order: Optional[int] = None
+ """
+ 話者の順番。
+
+ :attr:`SpeakerMeta.styles` は、この値に対して昇順に並んでいるべきである。
+ """
+
@pydantic.dataclasses.dataclass
class SpeakerMeta:
@@ -61,20 +68,13 @@ class SpeakerMeta:
version: StyleVersion
"""話者のUUID。"""
- speaker_order: Optional[int] = None
+ order: Optional[int] = None
"""
話者の順番。
``SpeakerMeta`` の列は、この値に対して昇順に並んでいるべきである。
"""
- style_order: List[StyleId] = dataclasses.field(default_factory=list)
- """
- 話者に属するスタイルの順番。
-
- :attr:`styles` はこの並びに沿うべきである。
- """
-
@pydantic.dataclasses.dataclass
class SupportedDevices: