ForcePilot/backend/package/yuxi/knowledge/parser/asr.py

"""ASR 解析器 — 将音频文件转写为 Markdown 文本。"""

import os
import tempfile
from pathlib import Path
from typing import Any

import httpx

from yuxi.knowledge.parser.base import BaseDocumentProcessor, DocumentProcessorException
from yuxi.knowledge.parser.ffmpeg import (
    FFMPEG_MAX_AUDIO_DURATION_SECS,
    MAX_AUDIO_BYTES,
    convert_to_wav,
    probe_audio_duration,
)


class ASRProcessor(BaseDocumentProcessor):
    """音频文件 ASR 解析器，将音频转写为 Markdown 文本。"""

    def get_service_name(self) -> str:
        return "asr"

    def get_supported_extensions(self) -> list[str]:
        return [".mp3", ".wav", ".m4a", ".flac", ".ogg", ".wma"]

    def process_file(self, file_path: str, params: dict[str, Any] | None = None) -> str:
        """
        解析音频文件，返回 Markdown 文本。

        流程:
        1. 检查文件大小和音频时长
        2. 预处理：转换为 WAV（16kHz, mono）
        3. ASR 转写
        4. 格式化为 Markdown（带时间戳段落）
        """
        params = params or {}

        # 文件大小检查
        file_size = Path(file_path).stat().st_size
        if file_size > MAX_AUDIO_BYTES:
            raise DocumentProcessorException(
                f"音频文件大小 {file_size / 1024 / 1024:.1f}MB 超过限制 {MAX_AUDIO_BYTES / 1024 / 1024:.0f}MB",
                service_name=self.get_service_name(),
            )

        # 时长检查
        duration = probe_audio_duration(file_path)
        if duration and duration > FFMPEG_MAX_AUDIO_DURATION_SECS:
            raise DocumentProcessorException(
                f"音频时长 {duration:.0f}s 超过限制 {FFMPEG_MAX_AUDIO_DURATION_SECS}s",
                service_name=self.get_service_name(),
            )

        # 预处理：转换为 WAV
        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
            wav_path = tmp.name
        try:
            convert_to_wav(file_path, wav_path)
            segments = self._transcribe(wav_path, params)
        finally:
            Path(wav_path).unlink(missing_ok=True)

        return self._segments_to_markdown(segments, file_path)

    def _transcribe(self, wav_path: str, params: dict) -> list[dict]:
        """ASR 转写，通过环境变量配置的 ASR 服务调用。"""
        asr_provider = params.get("asr_provider", os.getenv("ASR_PROVIDER", "funasr"))
        if asr_provider == "funasr":
            return self._transcribe_funasr(wav_path)
        elif asr_provider == "whisper":
            return self._transcribe_whisper(wav_path)
        else:
            raise DocumentProcessorException(
                f"不支持的 ASR 提供者: {asr_provider}",
                service_name=self.get_service_name(),
            )

    def _transcribe_funasr(self, wav_path: str) -> list[dict]:
        """调用 FunASR 服务进行转写。"""
        api_uri = os.getenv("FUNASR_API_URI", "http://funasr:10095")
        try:
            with open(wav_path, "rb") as f:
                with httpx.Client(timeout=120) as client:
                    resp = client.post(
                        f"{api_uri}/asr",
                        files={"file": (Path(wav_path).name, f, "audio/wav")},
                    )
                    resp.raise_for_status()
                    data = resp.json()

            # FunASR 返回格式适配
            # 常见格式: {"text": "...", "segments": [...]} 或 {"result": [...]}
            if isinstance(data, dict):
                if "segments" in data:
                    return data["segments"]
                # 单条结果包装为 segments
                text = data.get("text", data.get("result", ""))
                if isinstance(text, list):
                    return [{"start": 0, "end": 0, "text": t} for t in text]
                if text:
                    return [{"start": 0, "end": 0, "text": str(text)}]
            elif isinstance(data, list):
                return data

            return []
        except httpx.HTTPError as e:
            raise DocumentProcessorException(
                f"FunASR 服务调用失败: {e}",
                service_name=self.get_service_name(),
            )

    def _transcribe_whisper(self, wav_path: str) -> list[dict]:
        """调用 Whisper API（OpenAI 兼容）进行转写。"""
        api_key = os.getenv("WHISPER_API_KEY", "")
        base_url = os.getenv("WHISPER_BASE_URL", "https://api.openai.com/v1")
        if not api_key:
            raise DocumentProcessorException(
                "Whisper API Key 未配置 (WHISPER_API_KEY)",
                service_name=self.get_service_name(),
            )
        try:
            with open(wav_path, "rb") as f:
                with httpx.Client(timeout=120) as client:
                    resp = client.post(
                        f"{base_url}/audio/transcriptions",
                        headers={"Authorization": f"Bearer {api_key}"},
                        files={"file": (Path(wav_path).name, f, "audio/wav")},
                        data={"response_format": "verbose_json", "timestamp_granularities[]": "segment"},
                    )
                    resp.raise_for_status()
                    data = resp.json()

            segments = data.get("segments", [])
            if not segments and data.get("text"):
                return [{"start": 0, "end": 0, "text": data["text"]}]

            return [{"start": s.get("start", 0), "end": s.get("end", 0), "text": s.get("text", "")} for s in segments]
        except httpx.HTTPError as e:
            raise DocumentProcessorException(
                f"Whisper API 调用失败: {e}",
                service_name=self.get_service_name(),
            )

    @staticmethod
    def _segments_to_markdown(segments: list[dict], source: str) -> str:
        """将 ASR 片段格式化为 Markdown。"""
        lines = ["# 音频转写", "", f"> 来源: {source}", ""]
        for seg in segments:
            start = seg.get("start", 0)
            text = seg.get("text", "").strip()
            if text:
                mm_ss = f"{int(start) // 60:02d}:{int(start) % 60:02d}"
                lines.append(f"**[{mm_ss}]** {text}")
                lines.append("")
        return "\n".join(lines)

    def check_health(self) -> dict[str, Any]:
        """检查 ASR 服务健康状态。"""
        asr_provider = os.getenv("ASR_PROVIDER", "funasr")
        try:
            if asr_provider == "funasr":
                api_uri = os.getenv("FUNASR_API_URI", "http://funasr:10095")
                with httpx.Client(timeout=5) as client:
                    resp = client.get(f"{api_uri}/health")
                    if resp.status_code == 200:
                        return {"status": "healthy", "message": "FunASR 服务正常"}
                    return {"status": "unhealthy", "message": f"FunASR 返回 {resp.status_code}"}
            elif asr_provider == "whisper":
                if os.getenv("WHISPER_API_KEY"):
                    return {"status": "healthy", "message": "Whisper API 已配置"}
                return {"status": "unavailable", "message": "Whisper API Key 未配置"}
        except Exception as e:
            return {"status": "unhealthy", "message": str(e)}
        return {"status": "unavailable", "message": f"ASR 提供者 {asr_provider} 不可用"}