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>

api/app/features/conversation/ws/pipeline.py

@@ -2,7 +2,6 @@
 
 import asyncio
 import base64
-import io
 import time
 import uuid
 from dataclasses import dataclass, field
@@ -19,7 +18,7 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.agents.chat import ChatOrchestrator
 from app.agents.chat.reply_limits import segments_from_llm_response
-from app.core.agent_logging import agent_summary_enabled
+from app.core.agent_logging import agent_summary_enabled, log_asr_transcript_result
 from app.core.business_telemetry import business_span
 from app.core.config import settings
 from app.core.cos_url_keys import (
@@ -617,64 +616,6 @@ async def _delayed_listening_feedback(
     await _send_segment_transition_feedback(conversation_id, 0)
 
 
-# ── 长音频切片转写 ────────────────────────────────────────────
-
-MAX_ASR_CHUNK_MS = 55_000
-
-
-def _split_audio_bytes(audio_bytes: bytes, fmt: str) -> list[bytes]:
-    """用 pydub 将长音频按 ≤55 s 切片，每片导出为 16 kHz mono WAV（腾讯 ASR 3 MB 限制内）。"""
-    from pydub import AudioSegment as PydubSegment
-
-    audio = PydubSegment.from_file(io.BytesIO(audio_bytes), format=fmt)
-    duration_ms = len(audio)
-
-    if duration_ms <= MAX_ASR_CHUNK_MS:
-        return [audio_bytes]
-
-    mono_16k = audio.set_frame_rate(16000).set_channels(1).set_sample_width(2)
-    chunks: list[bytes] = []
-    for start in range(0, duration_ms, MAX_ASR_CHUNK_MS):
-        chunk = mono_16k[start : start + MAX_ASR_CHUNK_MS]
-        buf = io.BytesIO()
-        chunk.export(buf, format="wav")
-        chunks.append(buf.getvalue())
-    return chunks
-
-
-async def _transcribe_long_audio(audio_bytes: bytes, fmt: str = "m4a") -> str:
-    """超过 55 s 的音频自动切片后并行 ASR；短音频直接转写。"""
-    asr = get_asr_provider()
-    return await _transcribe_long_audio_inner(audio_bytes, fmt, asr)
-
-
-async def _transcribe_long_audio_inner(
-    audio_bytes: bytes, fmt: str, asr: Any
-) -> str:
-    try:
-        chunks = await asyncio.to_thread(_split_audio_bytes, audio_bytes, fmt)
-    except Exception as exc:
-        logger.warning("pydub 切片失败 ({}), 回退到直接转写", exc)
-        return await asr.transcribe(audio_bytes, format=fmt)
-
-    if len(chunks) <= 1:
-        return await asr.transcribe(audio_bytes, format=fmt)
-
-    logger.info("长音频切片: {} 段", len(chunks))
-    results = await asyncio.gather(
-        *[asr.transcribe(c, format="wav") for c in chunks],
-        return_exceptions=True,
-    )
-    texts: list[str] = []
-    for i, r in enumerate(results):
-        if isinstance(r, BaseException):
-            logger.warning("切片 {} 转写异常: {}", i, r)
-            continue
-        if r and not _is_transcribe_failure(r):
-            texts.append(r)
-    return "".join(texts)
-
-
 # ── 分段语音异步处理 ────────────────────────────────────────────
 
 
@@ -761,7 +702,19 @@ async def process_audio_segment(
                     segment_index,
                 )
             try:
-                transcript_text = await _transcribe_long_audio(audio_bytes, fmt="m4a")
+                asr = get_asr_provider()
+                transcript_text = await asr.transcribe(audio_bytes, format="m4a")
+                if transcript_text:
+                    log_asr_transcript_result(
+                        logger,
+                        text=transcript_text,
+                        conversation_id=conversation_id,
+                        voice_session_id=voice_session_id,
+                        segment_index=segment_index,
+                        duration_s=audio_duration,
+                        audio_len=len(audio_bytes),
+                        source="audio_segment",
+                    )
             except ASRTranscriptionError as e:
                 logger.warning(
                     "ASR 转写失败 segment_index={} conversation_id={}: {}",