ver0.1

app/services/baidu_speech.py

@@ -6,6 +6,7 @@ from typing import Any
 from aip import AipSpeech
 
 from app.config import Settings, settings as _default_settings
+from app.services.audio_wav import pcm_s16le_to_wav_bytes
 
 
 class BaiduSpeechNotConfiguredError(RuntimeError):
@@ -60,6 +61,31 @@ class BaiduSpeechService:
         merged["dev_pid"] = int(self._s.baidu_speech_asr_dev_pid)
         return self._client_or_raise().asr(speech, format, rate, merged)
 
+    def asr_16k_mono_pcm_or_wav_fallback(
+        self,
+        pcm_s16le: bytes,
+        *,
+        rate: int = 16000,
+        options: dict[str, Any] | None = None,
+    ) -> dict[str, Any]:
+        """先按 raw PCM 识别；若返回 err_no=3301（语音质量错误），再用 WAV 封装重试一次。
+
+        部分环境下 PCM 与 WAV 路径对边界样本表现不一致，重试可提高成功率。
+        """
+        r = self.asr(pcm_s16le, "pcm", rate, options)
+        if not isinstance(r, dict):
+            return r
+        if r.get("err_no") != 3301:
+            return r
+        if len(pcm_s16le) < 1000:
+            return r
+        try:
+            wav = pcm_s16le_to_wav_bytes(pcm_s16le, sample_rate=rate)
+        except Exception:
+            return r
+        r2 = self.asr(wav, "wav", rate, options)
+        return r2 if isinstance(r2, dict) else r
+
     def synthesis(
         self,
         text: str,