Pine
064b06ecf0
迁入 app/park：llm(对话)、rag(双路向量知识库)、tools/ai_tools(智能体工具)、 asr(语音识别)、s2s_bridge(实时语音桥)、vision_yolo/vision_llm(人脸/多模态)、 knowledge/*.md、vendor/s2s-cloud(s2s 云化栈)；routers 补 /api/ai|kb|asr|vision| s2s|tools 端点。智能体提示词/企业名录改读 park_config 主数据源。pyproject 加重依赖 (dashscope/numpy/openai/torch/transformers/ultralytics/websockets/soundfile/scipy/ nltk/jinja2)。TestClient 冒烟：ai/chat(无 Key 走本地规则)、kb、s2s、tools、display 均 200。
94 lines
3.0 KiB
Python
94 lines
3.0 KiB
Python
# -*- coding: utf-8 -*-
|
|
"""语音识别 —— 阿里云 DashScope paraformer-realtime-v2（流式）
|
|
前端录音上传（m4a/wav/ogg/opus/pcm）→ 这里转写为文本
|
|
"""
|
|
|
|
import logging
|
|
import os
|
|
import time
|
|
|
|
import certifi
|
|
|
|
# macOS 系统 Python 无系统 CA：必须在导入 dashscope 之前设置，
|
|
# aiohttp 才会使用 certifi 根证书（否则 WS 连接报 SSL 证书错误）
|
|
os.environ.setdefault("SSL_CERT_FILE", certifi.where())
|
|
|
|
import dashscope # noqa: E402
|
|
from dashscope.audio.asr import Recognition, RecognitionCallback # noqa: E402
|
|
|
|
from .config import settings # noqa: E402
|
|
|
|
log = logging.getLogger("dpm.asr")
|
|
|
|
|
|
def _err_text(result):
|
|
"""安全提取 RecognitionResult 的错误描述（其 __str__ 有缺陷）"""
|
|
for attr in ("message", "code"):
|
|
v = getattr(result, attr, None)
|
|
if v:
|
|
return f"{attr}={v}"
|
|
return repr(result)
|
|
|
|
# 前端 MediaRecorder 可能产生的容器/编码 → paraformer 格式名
|
|
FORMAT_ALIASES = {
|
|
"mp4": "m4a", "m4a": "m4a", "aac": "m4a",
|
|
"webm": "ogg", "opus": "opus", "ogg": "ogg",
|
|
"wav": "wav", "pcm": "pcm", "mp3": "mp3",
|
|
}
|
|
|
|
# m4a/mp3 等压缩格式采样率由服务端自动识别
|
|
_AUTO_SAMPLE_FORMATS = {"m4a", "mp3", "ogg", "opus", "wav"}
|
|
|
|
|
|
class _Callback(RecognitionCallback):
|
|
def __init__(self):
|
|
self.sentences = []
|
|
self.error = None
|
|
|
|
def on_message(self, message):
|
|
try:
|
|
header = message.get("header", {})
|
|
if header.get("action") != "result":
|
|
return
|
|
payload = message.get("payload", {})
|
|
sentences = payload.get("sentence", {}).get("sentences") or []
|
|
for s in sentences:
|
|
if s.get("sentence_end"):
|
|
self.sentences.append(s.get("text", ""))
|
|
except Exception as e: # noqa: BLE001
|
|
log.warning("ASR 消息解析异常: %s", e)
|
|
|
|
def on_error(self, result):
|
|
self.error = result
|
|
log.warning("ASR 错误: %s", _err_text(result))
|
|
|
|
|
|
def transcribe(audio_bytes: bytes, fmt: str = "m4a") -> str:
|
|
"""上传音频字节 → 返回转写文本（空串表示未识别到内容）"""
|
|
if not settings.DASHSCOPE_API_KEY:
|
|
raise RuntimeError("未配置 DASHSCOPE_API_KEY，无法使用阿里云语音识别")
|
|
|
|
dashscope.api_key = settings.DASHSCOPE_API_KEY
|
|
fmt = FORMAT_ALIASES.get((fmt or "m4a").lower().lstrip("."), "m4a")
|
|
|
|
cb = _Callback()
|
|
rec = Recognition(
|
|
model=settings.ASR_MODEL,
|
|
format=fmt,
|
|
sample_rate=16000 if fmt in ("pcm", "wav") else 0, # 压缩格式自动识别
|
|
callback=cb,
|
|
)
|
|
|
|
# 分块发送，模拟流式（适当间隔，避免触发服务端 batching 报错）
|
|
rec.start()
|
|
chunk = 16 * 1024
|
|
for i in range(0, len(audio_bytes), chunk):
|
|
rec.send_audio_frame(audio_bytes[i:i + chunk])
|
|
time.sleep(0.08)
|
|
time.sleep(0.5) # 等待尾部识别
|
|
rec.stop()
|
|
|
|
if cb.error:
|
|
raise RuntimeError(f"阿里云语音识别失败: {_err_text(cb.error)}")
|
|
return "".join(cb.sentences).strip()
|