feat(api): use Tencent ASR flash with 16k_zh_large and dev transcript logs
Replace CreateRecTask polling with recording-file flash API, add TENCENT_APP_ID, remove server-side pydub slicing, and log ASR recognition text at INFO in development. Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Kevin
@@ -1,7 +1,13 @@
|
||||
"""Tencent Cloud ASR adapter — implements ASRProvider port."""
|
||||
"""Tencent Cloud ASR adapter — 录音文件识别极速版 (Flash), implements ASRProvider port."""
|
||||
|
||||
import asyncio
|
||||
import base64
|
||||
import hashlib
|
||||
import hmac
|
||||
import json
|
||||
import time
|
||||
from typing import Any
|
||||
|
||||
import httpx
|
||||
|
||||
from app.core.business_telemetry import business_span
|
||||
from app.core.logging import get_logger
|
||||
@@ -9,82 +15,198 @@ from app.ports.asr import ASRTranscriptionError
|
||||
|
||||
logger = get_logger(__name__)
|
||||
|
||||
_FLASH_HOST = "asr.cloud.tencent.com"
|
||||
_FLASH_PATH_PREFIX = "/asr/flash/v1/"
|
||||
# 极速版本地上传上限（腾讯文档：≤100MB）
|
||||
_MAX_FLASH_AUDIO_BYTES = 100 * 1024 * 1024
|
||||
|
||||
|
||||
def _format_flash_sign_string(sorted_params: list[tuple[str, Any]]) -> str:
|
||||
"""与官方 flash_recognizer._format_sign_string 一致。"""
|
||||
signstr = f"POST{_FLASH_HOST}{_FLASH_PATH_PREFIX}"
|
||||
for key, value in sorted_params:
|
||||
if key == "appid":
|
||||
signstr += str(value)
|
||||
break
|
||||
signstr += "?"
|
||||
for key, value in sorted_params:
|
||||
if key == "appid":
|
||||
continue
|
||||
signstr += f"{key}={value}&"
|
||||
return signstr[:-1]
|
||||
|
||||
|
||||
def _build_flash_url_and_headers(
|
||||
secret_key: str, params: dict[str, Any]
|
||||
) -> tuple[str, dict[str, str]]:
|
||||
sorted_params = sorted(params.items(), key=lambda item: item[0])
|
||||
signstr = _format_flash_sign_string(sorted_params)
|
||||
signature = base64.b64encode(
|
||||
hmac.new(
|
||||
secret_key.encode("utf-8"),
|
||||
signstr.encode("utf-8"),
|
||||
hashlib.sha1,
|
||||
).digest()
|
||||
).decode("utf-8")
|
||||
url = "https://" + signstr[4:]
|
||||
headers = {
|
||||
"Host": _FLASH_HOST,
|
||||
"Authorization": signature,
|
||||
}
|
||||
return url, headers
|
||||
|
||||
|
||||
def _build_flash_query_params(
|
||||
*,
|
||||
app_id: str,
|
||||
secret_id: str,
|
||||
engine_type: str,
|
||||
voice_format: str,
|
||||
) -> dict[str, Any]:
|
||||
return {
|
||||
"appid": app_id,
|
||||
"secretid": secret_id,
|
||||
"timestamp": str(int(time.time())),
|
||||
"engine_type": engine_type,
|
||||
"voice_format": voice_format,
|
||||
"speaker_diarization": 0,
|
||||
"customization_id": "",
|
||||
"filter_dirty": 0,
|
||||
"filter_modal": 0,
|
||||
"filter_punc": 0,
|
||||
"convert_num_mode": 1,
|
||||
"word_info": 0,
|
||||
"first_channel_only": 1,
|
||||
"reinforce_hotword": 0,
|
||||
"sentence_max_length": 0,
|
||||
}
|
||||
|
||||
|
||||
class TencentASRProvider:
|
||||
def __init__(
|
||||
self,
|
||||
secret_id: str,
|
||||
secret_key: str,
|
||||
app_id: str,
|
||||
*,
|
||||
engine_type: str = "16k_zh_large",
|
||||
request_timeout_seconds: float = 60.0,
|
||||
):
|
||||
self._secret_id = secret_id
|
||||
self._secret_key = secret_key
|
||||
self._app_id = (app_id or "").strip()
|
||||
self._engine_type = engine_type
|
||||
self._client = None
|
||||
|
||||
def _get_client(self):
|
||||
if self._client is not None:
|
||||
return self._client
|
||||
try:
|
||||
from tencentcloud.asr.v20190614 import asr_client
|
||||
from tencentcloud.common import credential
|
||||
from tencentcloud.common.profile.client_profile import ClientProfile
|
||||
from tencentcloud.common.profile.http_profile import HttpProfile
|
||||
|
||||
cred = credential.Credential(self._secret_id, self._secret_key)
|
||||
http_profile = HttpProfile()
|
||||
http_profile.endpoint = "asr.tencentcloudapi.com"
|
||||
client_profile = ClientProfile()
|
||||
client_profile.httpProfile = http_profile
|
||||
self._client = asr_client.AsrClient(cred, "", client_profile)
|
||||
return self._client
|
||||
except Exception as e:
|
||||
logger.error("Tencent ASR client init failed: {}", e)
|
||||
return None
|
||||
self._request_timeout_seconds = request_timeout_seconds
|
||||
|
||||
def ensure_ready(self) -> bool:
|
||||
return bool(self._secret_id and self._secret_key and self._get_client())
|
||||
return bool(self._secret_id and self._secret_key and self._app_id)
|
||||
|
||||
async def transcribe(self, audio: bytes, format: str = "m4a") -> str:
|
||||
with business_span("asr.transcribe", provider="tencent"):
|
||||
with business_span("asr.transcribe", provider="tencent_flash"):
|
||||
return await self._transcribe_inner(audio, format)
|
||||
|
||||
async def _transcribe_inner(self, audio: bytes, format: str) -> str:
|
||||
client = self._get_client()
|
||||
if not client:
|
||||
if not self.ensure_ready():
|
||||
raise ASRTranscriptionError(
|
||||
"Tencent ASR client not initialized (check credentials)"
|
||||
"Tencent ASR flash not configured (need TENCENT_APP_ID, SECRET_ID, SECRET_KEY)"
|
||||
)
|
||||
if len(audio) > _MAX_FLASH_AUDIO_BYTES:
|
||||
raise ASRTranscriptionError(
|
||||
f"Audio exceeds {_MAX_FLASH_AUDIO_BYTES // (1024 * 1024)}MB flash upload limit"
|
||||
)
|
||||
|
||||
voice_format = (format or "m4a").lower()
|
||||
params = _build_flash_query_params(
|
||||
app_id=self._app_id,
|
||||
secret_id=self._secret_id,
|
||||
engine_type=self._engine_type,
|
||||
voice_format=voice_format,
|
||||
)
|
||||
url, headers = _build_flash_url_and_headers(self._secret_key, params)
|
||||
headers["Content-Type"] = "application/octet-stream"
|
||||
headers["Content-Length"] = str(len(audio))
|
||||
|
||||
try:
|
||||
from tencentcloud.asr.v20190614 import models
|
||||
|
||||
audio_base64 = base64.b64encode(audio).decode("utf-8")
|
||||
req = models.SentenceRecognitionRequest()
|
||||
req.EngSerViceType = self._engine_type
|
||||
req.SourceType = 1
|
||||
# 小写；与文档一致。iOS 常见为 m4a(AAC) 容器，与 16k 引擎匹配
|
||||
req.VoiceFormat = (format or "m4a").lower()
|
||||
req.Data = audio_base64
|
||||
req.DataLen = len(audio)
|
||||
|
||||
# 腾讯 SDK 为同步阻塞调用；放到线程池里避免卡住事件循环。
|
||||
resp = await asyncio.to_thread(client.SentenceRecognition, req)
|
||||
text = (resp.Result or "").strip()
|
||||
if text:
|
||||
return text
|
||||
err = getattr(resp, "Error", None) or getattr(resp, "Message", None)
|
||||
logger.warning(
|
||||
"Tencent ASR empty Result, audio_len={} format={} err={}",
|
||||
len(audio),
|
||||
req.VoiceFormat,
|
||||
err,
|
||||
)
|
||||
raise ASRTranscriptionError(
|
||||
"Tencent ASR empty Result (check sample rate / format / audio)"
|
||||
)
|
||||
async with httpx.AsyncClient() as client:
|
||||
resp = await client.post(
|
||||
url,
|
||||
headers=headers,
|
||||
content=audio,
|
||||
timeout=self._request_timeout_seconds,
|
||||
)
|
||||
if resp.status_code >= 400:
|
||||
raise ASRTranscriptionError(
|
||||
f"Tencent ASR flash HTTP {resp.status_code}: {resp.text[:200]}"
|
||||
)
|
||||
payload = resp.json()
|
||||
except ASRTranscriptionError:
|
||||
raise
|
||||
except httpx.HTTPError as e:
|
||||
logger.error("Tencent ASR flash HTTP failed: {}", e, exc_info=True)
|
||||
raise ASRTranscriptionError(f"Tencent ASR flash HTTP failed: {e!s}") from e
|
||||
except json.JSONDecodeError as e:
|
||||
logger.error("Tencent ASR flash invalid JSON: {}", e, exc_info=True)
|
||||
raise ASRTranscriptionError("Tencent ASR flash returned invalid JSON") from e
|
||||
except Exception as e:
|
||||
logger.error("Tencent ASR transcribe failed: {}", e, exc_info=True)
|
||||
raise ASRTranscriptionError(f"Tencent ASR transcribe failed: {e!s}") from e
|
||||
logger.error("Tencent ASR flash transcribe failed: {}", e, exc_info=True)
|
||||
raise ASRTranscriptionError(f"Tencent ASR flash transcribe failed: {e!s}") from e
|
||||
|
||||
return self._parse_flash_response(payload, audio_len=len(audio), voice_format=voice_format)
|
||||
|
||||
def _parse_flash_response(
|
||||
self, payload: dict[str, Any], *, audio_len: int, voice_format: str
|
||||
) -> str:
|
||||
code = payload.get("code")
|
||||
if code != 0:
|
||||
message = payload.get("message") or "unknown error"
|
||||
request_id = payload.get("request_id", "")
|
||||
logger.warning(
|
||||
"Tencent ASR flash error code={} message={} request_id={} audio_len={} format={}",
|
||||
code,
|
||||
message,
|
||||
request_id,
|
||||
audio_len,
|
||||
voice_format,
|
||||
)
|
||||
if code == 4004:
|
||||
raise ASRTranscriptionError(
|
||||
"Tencent ASR flash resource pack exhausted (purchase 录音文件识别极速版)"
|
||||
)
|
||||
if code == 4003:
|
||||
raise ASRTranscriptionError(
|
||||
"Tencent ASR flash service not enabled in console"
|
||||
)
|
||||
raise ASRTranscriptionError(
|
||||
f"Tencent ASR flash failed (code={code}): {message}"
|
||||
)
|
||||
|
||||
flash_result = payload.get("flash_result") or []
|
||||
texts: list[str] = []
|
||||
for channel in flash_result:
|
||||
if not isinstance(channel, dict):
|
||||
continue
|
||||
text = (channel.get("text") or "").strip()
|
||||
if text:
|
||||
texts.append(text)
|
||||
combined = "".join(texts)
|
||||
if combined:
|
||||
logger.debug(
|
||||
"Tencent ASR flash ok request_id={} audio_len={} audio_duration_ms={} "
|
||||
"voice_format={} chars={}",
|
||||
payload.get("request_id"),
|
||||
audio_len,
|
||||
payload.get("audio_duration"),
|
||||
voice_format,
|
||||
len(combined),
|
||||
)
|
||||
return combined
|
||||
|
||||
logger.warning(
|
||||
"Tencent ASR flash empty flash_result, audio_len={} format={} request_id={}",
|
||||
audio_len,
|
||||
voice_format,
|
||||
payload.get("request_id"),
|
||||
)
|
||||
raise ASRTranscriptionError(
|
||||
"Tencent ASR flash empty result (check sample rate / format / audio)"
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user