feat!: Rust APIだけTextAnalyzerをパブリックにする

VOICEVOX · Jan 5, 2025 · 50645d8 · 50645d8
1 parent 6ff13e0
commit 50645d8
Show file tree

Hide file tree

Showing 25 changed files with 230 additions and 188 deletions.
diff --git a/crates/voicevox_core/src/__internal/doctest_fixtures.rs b/crates/voicevox_core/src/__internal/doctest_fixtures.rs
@@ -22,7 +22,7 @@ pub async fn synthesizer_with_sample_voice_model(
         #[cfg(feature = "link-onnxruntime")]
         crate::nonblocking::Onnxruntime::init_once().await?,
     )
-    .open_jtalk(crate::nonblocking::OpenJtalk::new(open_jtalk_dic_dir).await?)
+    .text_analyzer(crate::nonblocking::OpenJtalk::new(open_jtalk_dic_dir).await?)
     .acceleration_mode(AccelerationMode::Cpu)
     .build()?;
 

diff --git a/crates/voicevox_core/src/blocking.rs b/crates/voicevox_core/src/blocking.rs
@@ -3,7 +3,8 @@
 pub use crate::{
     engine::open_jtalk::blocking::OpenJtalk, infer::runtimes::onnxruntime::blocking::Onnxruntime,
     synthesizer::blocking::AudioFeature, synthesizer::blocking::Synthesizer,
-    user_dict::dict::blocking::UserDict, voice_model::blocking::VoiceModelFile,
+    text_analyzer::blocking::TextAnalyzer, user_dict::dict::blocking::UserDict,
+    voice_model::blocking::VoiceModelFile,
 };
 
 pub mod onnxruntime {

diff --git a/crates/voicevox_core/src/engine/full_context_label.rs b/crates/voicevox_core/src/engine/full_context_label.rs
@@ -7,7 +7,6 @@ use crate::{
 use jlabel::Label;
 use smallvec::SmallVec;
 
-// FIXME: 入力テキストをここで持って、メッセージに含む
 #[derive(thiserror::Error, Debug)]
 #[error("入力テキストからのフルコンテキストラベル抽出に失敗しました: {context}")]
 pub(crate) struct FullContextLabelError {
@@ -428,7 +427,7 @@ mod tests {
         let open_jtalk = crate::nonblocking::OpenJtalk::new(OPEN_JTALK_DIC_DIR)
             .await
             .unwrap();
-        assert_eq!(&open_jtalk.extract_fullcontext(text).unwrap(), labels);
+        assert_eq!(&open_jtalk.0.extract_fullcontext(text).unwrap(), labels);
     }
 
     #[apply(label_cases)]
@@ -451,7 +450,7 @@ mod tests {
             .await
             .unwrap();
         assert_eq!(
-            &extract_full_context_label(&open_jtalk, text).unwrap(),
+            &extract_full_context_label(&open_jtalk.0, text).unwrap(),
             accent_phrase
         );
     }

diff --git a/crates/voicevox_core/src/engine/kana_parser.rs b/crates/voicevox_core/src/engine/kana_parser.rs
@@ -121,6 +121,9 @@ fn text_to_accent_phrase(phrase: &str) -> KanaParseResult<AccentPhrase> {
 
 pub(crate) fn parse_kana(text: &str) -> KanaParseResult<Vec<AccentPhrase>> {
     const TERMINATOR: char = '\0';
+    if text.is_empty() {
+        return Ok(vec![]);
+    }
     let mut parsed_result = Vec::new();
     let chars_of_text = text.chars().chain([TERMINATOR]);
     let mut phrase = String::new();

diff --git a/crates/voicevox_core/src/engine/mod.rs b/crates/voicevox_core/src/engine/mod.rs
@@ -10,9 +10,8 @@ pub(crate) mod open_jtalk;
 pub(crate) use self::acoustic_feature_extractor::OjtPhoneme;
 pub(crate) use self::audio_file::to_s16le_pcm;
 pub use self::audio_file::wav_from_s16le;
-pub(crate) use self::full_context_label::{extract_full_context_label, FullContextLabelError};
+pub(crate) use self::full_context_label::extract_full_context_label;
 pub(crate) use self::interpret_query::{initial_process, split_mora, DecoderFeature};
 pub(crate) use self::kana_parser::{create_kana, parse_kana, KanaParseError};
 pub use self::model::{AccentPhrase, AudioQuery, Mora};
 pub(crate) use self::mora_list::mora2text;
-pub use self::open_jtalk::FullcontextExtractor;
diff --git a/crates/voicevox_core/src/engine/open_jtalk.rs b/crates/voicevox_core/src/engine/open_jtalk.rs
@@ -21,7 +21,7 @@ pub(crate) struct OpenjtalkFunctionError {
     source: Option<Text2MecabError>,
 }
 
-pub trait FullcontextExtractor: Clone + Send + Sync + 'static {
+pub(crate) trait FullcontextExtractor {
     fn extract_fullcontext(&self, text: &str) -> anyhow::Result<Vec<String>>;
 }
 
@@ -38,7 +38,10 @@ pub(crate) mod blocking {
 
     use crate::error::ErrorRepr;
 
-    use super::{FullcontextExtractor, OpenjtalkFunctionError};
+    use super::{
+        super::{extract_full_context_label, AccentPhrase},
+        FullcontextExtractor, OpenjtalkFunctionError,
+    };
 
     /// テキスト解析器としてのOpen JTalk。
     #[derive(Clone)]
@@ -80,12 +83,19 @@ pub(crate) mod blocking {
     }
 
     impl FullcontextExtractor for self::OpenJtalk {
+        fn extract_fullcontext(&self, text: &str) -> anyhow::Result<Vec<String>> {
+            self.0.extract_fullcontext(text)
+        }
+    }
+
+    // TODO: このコードの移動
+    impl FullcontextExtractor for Inner {
         fn extract_fullcontext(&self, text: &str) -> anyhow::Result<Vec<String>> {
             let Resources {
                 mecab,
                 njd,
                 jpcommon,
-            } = &mut *self.0.resources.lock().unwrap();
+            } = &mut *self.resources.lock().unwrap();
 
             jpcommon.refresh();
             njd.refresh();
@@ -129,6 +139,15 @@ pub(crate) mod blocking {
         }
     }
 
+    impl crate::blocking::TextAnalyzer for self::OpenJtalk {
+        fn analyze(&self, text: &str) -> anyhow::Result<Vec<AccentPhrase>> {
+            if text.is_empty() {
+                return Ok(Vec::new());
+            }
+            Ok(extract_full_context_label(&*self.0, text)?)
+        }
+    }
+
     pub(super) struct Inner {
         resources: std::sync::Mutex<Resources>,
         dict_dir: Utf8PathBuf,
@@ -196,7 +215,7 @@ pub(crate) mod blocking {
 pub(crate) mod nonblocking {
     use camino::Utf8Path;
 
-    use super::FullcontextExtractor;
+    use super::super::{extract_full_context_label, AccentPhrase};
 
     /// テキスト解析器としてのOpen JTalk。
     ///
@@ -207,7 +226,7 @@ pub(crate) mod nonblocking {
     /// [blocking]: https://docs.rs/crate/blocking
     /// [`nonblocking`モジュールのドキュメント]: crate::nonblocking
     #[derive(Clone)]
-    pub struct OpenJtalk(super::blocking::OpenJtalk);
+    pub struct OpenJtalk(pub(in super::super) super::blocking::OpenJtalk);
 
     impl self::OpenJtalk {
         pub async fn new(open_jtalk_dict_dir: impl AsRef<Utf8Path>) -> crate::result::Result<Self> {
@@ -231,9 +250,16 @@ pub(crate) mod nonblocking {
         }
     }
 
-    impl FullcontextExtractor for self::OpenJtalk {
-        fn extract_fullcontext(&self, text: &str) -> anyhow::Result<Vec<String>> {
-            self.0.extract_fullcontext(text)
+    impl crate::nonblocking::TextAnalyzer for self::OpenJtalk {
+        async fn analyze(&self, text: &str) -> anyhow::Result<Vec<AccentPhrase>> {
+            if text.is_empty() {
+                return Ok(Vec::new());
+            }
+            let inner = self.0 .0.clone();
+            let text = text.to_owned();
+            crate::task::asyncify(move || extract_full_context_label(&*inner, &text))
+                .await
+                .map_err(Into::into)
         }
     }
 }
@@ -345,7 +371,7 @@ mod tests {
         let open_jtalk = super::nonblocking::OpenJtalk::new(OPEN_JTALK_DIC_DIR)
             .await
             .unwrap();
-        let result = open_jtalk.extract_fullcontext(text);
+        let result = open_jtalk.0.extract_fullcontext(text);
         assert_debug_fmt_eq!(expected, result);
     }
 
@@ -360,7 +386,7 @@ mod tests {
             .await
             .unwrap();
         for _ in 0..10 {
-            let result = open_jtalk.extract_fullcontext(text);
+            let result = open_jtalk.0.extract_fullcontext(text);
             assert_debug_fmt_eq!(expected, result);
         }
     }

diff --git a/crates/voicevox_core/src/error.rs b/crates/voicevox_core/src/error.rs
@@ -1,8 +1,6 @@
 use crate::{
-    devices::DeviceAvailabilities,
-    engine::{FullContextLabelError, KanaParseError},
-    user_dict::InvalidWordError,
-    StyleId, StyleType, VoiceModelId,
+    devices::DeviceAvailabilities, engine::KanaParseError, user_dict::InvalidWordError, StyleId,
+    StyleType, VoiceModelId,
 };
 //use engine::
 use duplicate::duplicate_item;
@@ -19,7 +17,6 @@ pub struct Error(#[from] ErrorRepr);
 #[duplicate_item(
     E;
     [ LoadModelError ];
-    [ FullContextLabelError ];
     [ KanaParseError ];
     [ InvalidWordError ];
 )]
@@ -48,7 +45,7 @@ impl Error {
             ErrorRepr::StyleNotFound { .. } => ErrorKind::StyleNotFound,
             ErrorRepr::ModelNotFound { .. } => ErrorKind::ModelNotFound,
             ErrorRepr::RunModel { .. } => ErrorKind::RunModel,
-            ErrorRepr::ExtractFullContextLabel(_) => ErrorKind::ExtractFullContextLabel,
+            ErrorRepr::AnalyzeText { .. } => ErrorKind::AnalyzeText,
             ErrorRepr::ParseKana(_) => ErrorKind::ParseKana,
             ErrorRepr::LoadUserDict(_) => ErrorKind::LoadUserDict,
             ErrorRepr::SaveUserDict(_) => ErrorKind::SaveUserDict,
@@ -99,8 +96,12 @@ pub(crate) enum ErrorRepr {
     #[error("推論に失敗しました")]
     RunModel(#[source] anyhow::Error),
 
-    #[error(transparent)]
-    ExtractFullContextLabel(#[from] FullContextLabelError),
+    #[error("入力テキストの解析に失敗しました")]
+    AnalyzeText {
+        text: String,
+        #[source]
+        source: anyhow::Error,
+    },
 
     #[error(transparent)]
     ParseKana(#[from] KanaParseError),
@@ -150,8 +151,8 @@ pub enum ErrorKind {
     ModelNotFound,
     /// 推論に失敗した。
     RunModel,
-    /// コンテキストラベル出力に失敗した。
-    ExtractFullContextLabel,
+    /// 入力テキストの解析に失敗した。
+    AnalyzeText,
     /// AquesTalk風記法のテキストの解析に失敗した。
     ParseKana,
     /// ユーザー辞書を読み込めなかった。

diff --git a/crates/voicevox_core/src/lib.rs b/crates/voicevox_core/src/lib.rs
@@ -84,7 +84,7 @@ use rstest_reuse;
 
 pub use self::{
     devices::SupportedDevices,
-    engine::{wav_from_s16le, AccentPhrase, AudioQuery, FullcontextExtractor, Mora},
+    engine::{wav_from_s16le, AccentPhrase, AudioQuery, Mora},
     error::{Error, ErrorKind},
     metas::{
         RawStyleId, RawStyleVersion, SpeakerMeta, StyleId, StyleMeta, StyleType, StyleVersion,

diff --git a/crates/voicevox_core/src/nonblocking.rs b/crates/voicevox_core/src/nonblocking.rs
@@ -20,7 +20,8 @@
 pub use crate::{
     engine::open_jtalk::nonblocking::OpenJtalk,
     infer::runtimes::onnxruntime::nonblocking::Onnxruntime, synthesizer::nonblocking::Synthesizer,
-    user_dict::dict::nonblocking::UserDict, voice_model::nonblocking::VoiceModelFile,
+    text_analyzer::nonblocking::TextAnalyzer, user_dict::dict::nonblocking::UserDict,
+    voice_model::nonblocking::VoiceModelFile,
 };
 
 pub mod onnxruntime {