[Wake/VAD/STT]：完成本地唤醒、人声端点检测与转写入口，包含小杰小杰唤醒、VAD 录音切分、Metadata STT 和 sherpa-onnx 错误边界

2026-06-17 18:18:37 +08:00
parent 90ba2bf3b0
commit 4d6232ed29
6 changed files with 389 additions and 5 deletions
@@ -0,0 +1,53 @@
+from __future__ import annotations
+
+from .models import AudioFrame, ErrorCode, ProviderError, WakeEvent
+
+
+class KeywordWakeWordProvider:
+    def __init__(self, keyword: str = "小杰小杰", threshold: float = 0.5) -> None:
+        self.keyword = keyword
+        self.threshold = threshold
+        self.loaded = False
+
+    def load(self) -> None:
+        self.loaded = True
+
+    def detect(self, frame: AudioFrame) -> WakeEvent | None:
+        if not self.loaded:
+            raise ProviderError(
+                ErrorCode.WAKE_MODEL_LOAD_FAILED,
+                "wakeword provider is not loaded",
+                False,
+                "keyword-wakeword",
+                "wakeword",
+            )
+        metadata = frame.metadata
+        confidence = float(metadata.get("wake_confidence", 1.0 if metadata.get("wake") else 0.0))
+        phrase = str(metadata.get("wake_word", metadata.get("text", "")))
+        matched = bool(metadata.get("wake")) or phrase.strip() == self.keyword
+        if matched and confidence >= self.threshold:
+            return WakeEvent(self.keyword, confidence, frame.timestamp_ms)
+        return None
+
+    def reset(self) -> None:
+        return None
+
+
+class MissingWakeWordModelProvider:
+    def __init__(self, model_path: str) -> None:
+        self.model_path = model_path
+
+    def load(self) -> None:
+        raise ProviderError(
+            ErrorCode.WAKE_MODEL_MISSING,
+            f"wakeword model is missing: {self.model_path}",
+            False,
+            "wakeword-model",
+            "wakeword",
+        )
+
+    def detect(self, frame: AudioFrame) -> WakeEvent | None:
+        return None
+
+    def reset(self) -> None:
+        return None