Pass VoiceDesign caption through StackChan API

README.md

@@ -31,6 +31,7 @@ Parameters:
 - `duration_scale`: optional duration predictor scale. Default is `0.95`
 - `steps`: diffusion steps. Default is `18`
 - `seed`: optional. Leave empty for stable per-speaker seed
+- `caption`: optional VoiceDesign prompt. Defaults to the StackChan voice style
 
 Example:
 
@@ -48,8 +49,10 @@ Response shape:
   "seed": "4683784571875087395",
   "durationScale": 0.95,
   "metadata": {
+    "checkpoint_repo": "Aratako/Irodori-TTS-600M-v3-VoiceDesign",
     "reference_used": true,
-    "reference_path": "reference/stackchan-reference.mp3"
+    "reference_path": "reference/stackchan-reference.mp3",
+    "caption_used": true
   },
   "mp3StreamingUrl": "https://...",
   "mp3DownloadUrl": "https://...",
@@ -80,6 +83,7 @@ Configuration:
 - `DEFAULT_DURATION_SCALE`: default `0.95`
 - `DEFAULT_STEPS`: default `18`
 - `DEFAULT_SEED`: default `3407`
+- `DEFAULT_CAPTION`: default StackChan VoiceDesign prompt
 - `MAX_TEXT_LENGTH`: default `240`
 
 ## Cost

app.py

@@ -22,6 +22,10 @@ DEFAULT_SECONDS = "" if DEFAULT_SECONDS_RAW.lower() in {"", "auto", "none"} else
 DEFAULT_DURATION_SCALE = float(os.getenv("DEFAULT_DURATION_SCALE", "0.95"))
 DEFAULT_STEPS = int(os.getenv("DEFAULT_STEPS", "18"))
 DEFAULT_SEED = int(os.getenv("DEFAULT_SEED", "3407"))
+DEFAULT_CAPTION = os.getenv(
+    "DEFAULT_CAPTION",
+    "若く元気な女性の声。近い距離感で、明るくやわらかく自然に話している。",
+)
 
 OUTPUT_DIR = Path(os.getenv("OUTPUT_DIR", "/tmp/stackchan-audio"))
 OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
@@ -113,6 +117,7 @@ def synthesis(
     duration_scale: float = Query(default=DEFAULT_DURATION_SCALE, gt=0.0, le=2.0),
     steps: int = Query(default=DEFAULT_STEPS, ge=1, le=80),
     seed: str = Query(default=""),
+    caption: str = Query(default=DEFAULT_CAPTION),
 ) -> JSONResponse:
     _check_key(key)
     text = text.strip()
@@ -130,6 +135,7 @@ def synthesis(
             duration_scale=float(duration_scale),
             steps=int(steps),
             seed=seed_value,
+            caption=str(caption).strip() or DEFAULT_CAPTION,
             api_name="/synthesize",
         )
         metadata = _result_metadata(result)