fix: bug

crates/sbv2_core/src/tts_util.rs

@@ -2,6 +2,8 @@ use std::io::Cursor;
 
 use crate::error::Result;
 use crate::jtalk::JTalkProcess;
+use crate::mora::MORA_KATA_TO_MORA_PHONEMES;
+use crate::norm::PUNCTUATIONS;
 use crate::{jtalk, nlp, norm, tokenizer, utils};
 use hound::{SampleFormat, WavSpec, WavWriter};
 use ndarray::{concatenate, s, Array, Array1, Array2, Array3, Axis};
@@ -111,6 +113,7 @@ pub fn parse_text_blocking(
     if let Some(given_tones) = given_tones {
         tones = given_tones;
     }
+    println!("tones: {:?}", tones);
     let (phones, tones, lang_ids) = nlp::cleaned_text_to_sequence(phones, tones);
 
     let phones = utils::intersperse(&phones, 0);
@@ -188,3 +191,23 @@ pub fn array_to_vec(audio_array: Array3<f32>) -> Result<Vec<u8>> {
     writer.finalize()?;
     Ok(cursor.into_inner())
 }
+
+pub fn kata_tone2phone_tone(kata_tone: Vec<(String, i32)>) -> Vec<(String, i32)> {
+    let mut results = vec![("_".to_string(), 0)];
+    for (mora, tone) in kata_tone {
+        if PUNCTUATIONS.contains(&mora.as_str()) {
+            results.push((mora, 0));
+            continue;
+        } else {
+            let (consonant, vowel) = MORA_KATA_TO_MORA_PHONEMES.get(&mora).unwrap();
+            if let Some(consonant) = consonant {
+                results.push((consonant.to_string(), tone));
+                results.push((vowel.to_string(), tone));
+            } else {
+                results.push((vowel.to_string(), tone));
+            }
+        }
+    }
+    results.push(("_".to_string(), 0));
+    results
+}

crates/sbv2_editor/src/main.rs

@@ -6,6 +6,7 @@ use axum::{
     Json, Router,
     http::header::CONTENT_TYPE,
 };
+use sbv2_core::tts_util::kata_tone2phone_tone;
 use sbv2_core::{jtalk::JTalk, tts::{TTSModelHolder, SynthesizeOptions}, tts_util::preprocess_parse_text};
 use serde::{Deserialize, Serialize};
 use tokio::{fs, net::TcpListener, sync::Mutex};
@@ -68,9 +69,16 @@ async fn synthesis(
     State(state): State<AppState>,
     Json(request): Json<RequestSynthesis>,
 ) -> AppResult<impl IntoResponse> {
-    let mut tones: Vec<i32> = request.audio_query.iter().map(|query| query.tone).collect();
-    tones.insert(0, 0);
-    tones.push(0);
+    let phone_tone = request
+        .audio_query
+        .iter()
+        .map(|query| (query.kana.clone(), query.tone))
+        .collect::<Vec<_>>();
+    let phone_tone = kata_tone2phone_tone(phone_tone);
+    let tones = phone_tone
+        .iter()
+        .map(|(_, tone)| *tone)
+        .collect::<Vec<_>>();
     let buffer = {
         let mut tts_model = state.tts_model.lock().await;
         tts_model.easy_synthesize_neo(