metas出力時に話者情報をマージする

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};

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())
     }
 }
 

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 {

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() {