feat: Support markup input for Cloud TTS Chirp 3: HD voice synthesis

Google APIs · copybara-github · commit 8f7ef1c0eed3 · 2025-05-05T07:08:19.000-07:00
feat: Support pinyin/yomigana custom pronunciation encodings for cmn-cn/ja-jp

PiperOrigin-RevId: 754921874
diff --git a/google/cloud/texttospeech/v1/cloud_tts.proto b/google/cloud/texttospeech/v1/cloud_tts.proto
@@ -191,6 +191,29 @@ message CustomPronunciationParams {
     // X-SAMPA, such as apple -> "{p@l".
     // https://en.wikipedia.org/wiki/X-SAMPA
     PHONETIC_ENCODING_X_SAMPA = 2;
+
+    // For reading-to-pron conversion to work well, the `pronunciation` field
+    //  should only contain Kanji, Hiragana, and Katakana.
+    //
+    // The pronunciation can also contain pitch accents.
+    // The start of a pitch phrase is specified with `^` and the down-pitch
+    // position is specified with `!`, for example:
+    //
+    //     phrase:端  pronunciation:^はし
+    //     phrase:箸  pronunciation:^は!し
+    //     phrase:橋  pronunciation:^はし!
+    //
+    // We currently only support the Tokyo dialect, which allows at most one
+    // down-pitch per phrase (i.e. at most one `!` between `^`).
+    PHONETIC_ENCODING_JAPANESE_YOMIGANA = 3;
+
+    // Used to specify pronunciations for Mandarin words. See
+    // https://en.wikipedia.org/wiki/Pinyin.
+    //
+    // For example: 朝阳, the pronunciation is "chao2 yang2". The number
+    // represents the tone, and there is a space between syllables. Neutral
+    // tones are represented by 5, for example 孩子 "hai2 zi5".
+    PHONETIC_ENCODING_PINYIN = 4;
   }
 
   // The phrase to which the customization is applied.
@@ -238,6 +261,10 @@ message SynthesisInput {
     // The raw text to be synthesized.
     string text = 1;
 
+    // Markup for HD voices specifically. This field may not be used with any
+    // other voices.
+    string markup = 5;
+
     // The SSML document to be synthesized. The SSML document must be valid
     // and well-formed. Otherwise the RPC will fail and return
     // [google.rpc.Code.INVALID_ARGUMENT][google.rpc.Code.INVALID_ARGUMENT]. For
@@ -453,6 +480,10 @@ message StreamingSynthesisInput {
     // contains complete, terminating sentences, which results in better prosody
     // in the output audio.
     string text = 1;
+
+    // Markup for HD voices specifically. This field may not be used with any
+    // other voices.
+    string markup = 5;
   }
 }
 
