98 lines
3.5 KiB
Python
98 lines
3.5 KiB
Python
from __future__ import annotations
|
||
|
||
from threading import Lock
|
||
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):
|
||
"""未配置 `BAIDU_APP_ID` / `BAIDU_API_KEY` / `BAIDU_SECRET_KEY` 时调用接口会抛出。"""
|
||
|
||
|
||
class BaiduSpeechService:
|
||
"""百度短语音识别（asr）与在线语音合成（synthesis），基于 `baidu-aip` 的 `AipSpeech`。"""
|
||
|
||
def __init__(self, app_settings: Settings | None = None) -> None:
|
||
self._s = app_settings or _default_settings
|
||
self._client: AipSpeech | None = None
|
||
self._lock = Lock()
|
||
|
||
@property
|
||
def configured(self) -> bool:
|
||
return self._s.baidu_speech_configured
|
||
|
||
def _client_or_raise(self) -> AipSpeech:
|
||
if not self.configured:
|
||
raise BaiduSpeechNotConfiguredError(
|
||
"百度语音未配置：请设置 BAIDU_APP_ID、BAIDU_API_KEY、BAIDU_SECRET_KEY。"
|
||
)
|
||
with self._lock:
|
||
if self._client is None:
|
||
client = AipSpeech(
|
||
self._s.baidu_speech_app_id,
|
||
self._s.baidu_speech_api_key,
|
||
self._s.baidu_speech_secret_key,
|
||
)
|
||
if self._s.baidu_speech_connection_timeout_ms is not None:
|
||
client.setConnectionTimeoutInMillis(
|
||
self._s.baidu_speech_connection_timeout_ms
|
||
)
|
||
if self._s.baidu_speech_socket_timeout_ms is not None:
|
||
client.setSocketTimeoutInMillis(self._s.baidu_speech_socket_timeout_ms)
|
||
self._client = client
|
||
return self._client
|
||
|
||
def asr(
|
||
self,
|
||
speech: bytes | None = None,
|
||
format: str = "pcm",
|
||
rate: int = 16000,
|
||
options: dict[str, Any] | None = None,
|
||
) -> dict[str, Any]:
|
||
"""短语音识别。返回百度 JSON（含 `err_no`、`result` 等）。
|
||
|
||
固定使用普通话模型（`dev_pid` 来自配置），避免未传参时误用服务端默认导致偏英语等结果。
|
||
"""
|
||
merged: dict[str, Any] = dict(options or {})
|
||
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,
|
||
lang: str = "zh",
|
||
ctp: int = 1,
|
||
options: dict[str, Any] | None = None,
|
||
) -> bytes | dict[str, Any]:
|
||
"""在线语音合成。成功为音频二进制；失败为错误信息 dict。"""
|
||
return self._client_or_raise().synthesis(text, lang, ctp, options)
|