add sentences borders enable in pc player

sentence_api/processing.py

@@ -14,7 +14,8 @@ from .audio_metrics import (
     refine_sentence_start_ms,
 )
 from .config import Settings
-from .generate_boundaries import ALGORITHM_VERSION, make_entry
+from .generate_boundaries import ALGORITHM_VERSION, make_entry, media_duration_ms
+from sentence_analysis import detect_sentence_boundaries
 from .models import SentenceBoundary, SentenceBoundaryDocument
 from .oss import VolcanoOSSUploader
 from .repository import VideoRepository
@@ -27,6 +28,7 @@ from .transcription import (
 
 logger = logging.getLogger(__name__)
 MOSS_ALGORITHM_VERSION = "moss-punctuation-v3"
+SILENCE_ALGORITHM_VERSION = "silence-rms-v1"
 
 
 class VideoProcessor:
@@ -417,6 +419,67 @@ def document_from_transcript(
     )
 
 
+def generate_audio_boundary_document(
+    *,
+    video_hash: str,
+    audio_path: Path,
+    transcriber: Transcriber,
+    language: Optional[str] = None,
+) -> SentenceBoundaryDocument:
+    """Generate sentence boundaries for a client-supplied audio file.
+
+    The caller owns ``audio_path`` and is responsible for deleting it. Nothing
+    in this function copies or registers the audio in application storage.
+    """
+    duration_ms = media_duration_ms(audio_path)
+    transcript = None
+    if transcriber.available:
+        try:
+            transcript = transcriber.transcribe(audio_path, language)
+        except Exception as exc:
+            logger.warning("Audio boundary transcription failed; using silence fallback: %s", exc)
+
+    if transcript is not None:
+        document = document_from_transcript(
+            video_hash=video_hash,
+            duration_ms=duration_ms,
+            transcript=transcript,
+            language=language,
+            audio_path=audio_path,
+        )
+        if document.sentences:
+            return document
+
+    starts = detect_sentence_boundaries(audio_path)
+    if not starts:
+        raise ValueError("No sentence boundaries could be detected in the audio.")
+
+    starts_ms = sorted({max(0, min(duration_ms - 1, int(round(start * 1000)))) for start in starts})
+    sentences: List[SentenceBoundary] = []
+    for index, start_ms in enumerate(starts_ms):
+        end_ms = starts_ms[index + 1] if index + 1 < len(starts_ms) else duration_ms
+        if end_ms <= start_ms:
+            continue
+        sentences.append(
+            SentenceBoundary(
+                index=len(sentences),
+                start_ms=start_ms,
+                end_ms=end_ms,
+                text=None,
+                language=language,
+            )
+        )
+    if not sentences:
+        raise ValueError("Detected audio boundaries do not form valid sentence ranges.")
+
+    return SentenceBoundaryDocument(
+        video_hash=video_hash,
+        duration_ms=duration_ms,
+        algorithm_version=SILENCE_ALGORITHM_VERSION,
+        sentences=sentences,
+    )
+
+
 def _media_duration_ms(path: Path) -> int:
     from .generate_boundaries import media_duration_ms