Trtyr's Blog

1
~/bilibili-tools/bilibili-fetch/
2
├── scripts/
3
│   ├── fetch_subtitle.py    # 拉字幕
4
│   └── download_audio.py    # 下载音频
5
└── .venv/                   # 独立 Python 环境

1
cd ~/bilibili-tools/bilibili-fetch
2
uv venv .venv                                       # 或 python -m venv .venv
3
uv pip install --python .venv bilibili-api-python httpx

1
[credential]
2
sessdata = "你的 SESSDATA"
3
bili_jct = "你的 bili_jct"
4
buvid3 = "你的 buvid3"
5
buvid4 = "你的 buvid4"
6
dedeuserid = "你的 DedeUserID"
7
ac_time_value = ""   # 浏览器 cookie 里没有这项，留空即可，不影响拉字幕和下音频

1
#!/usr/bin/env python3
2
"""从B站视频拉取官方字幕。
3

4
输出契约 (stdout/stderr 分离, 便于调用方机械判断):
5
- 有字幕 → 字幕正文写入输出文件, stdout 输出文件绝对路径。
6
- 无字幕 → stdout 输出 NO_SUBTITLE, 不写文件。
7
"""
8
from __future__ import annotations
9

10
import argparse
11
import asyncio
12
import re
13
import sys
14
import tomllib
15
from pathlib import Path
16

17
import httpx
18
from bilibili_api import Credential, video
19

20
BVID_RE = re.compile(r"BV[1-9A-HJ-NP-Za-km-z]{10}")
21
SHORT_LINK_RE = re.compile(r"^https?://(?:www\.)?(?:b23\.tv|bili2233\.cn)/")
22
NO_SUBTITLE_SENTINEL = "NO_SUBTITLE"
23

24
CONFIG_PATH = Path.home() / ".config" / "bilibili-subtitle-fetch" / "config.toml"
25
DEFAULT_LANG = "zh-CN"
26
UA = (
27
    "Mozilla/5.0 (Windows NT 10.0; Win64; x64) "
28
    "AppleWebKit/537.36 (KHTML, like Gecko) Chrome/137.0.0.0 Safari/537.36"
29
)
30

31

32
def resolve_bvid(text: str) -> str:
33
    """BVID / 视频 URL / b23.tv 短链 → BV号。"""
34
    text = text.strip()
35
    m = BVID_RE.search(text)
36
    if m:
37
        return m.group(0)
38
    if SHORT_LINK_RE.match(text):
39
        resp = httpx.get(text, follow_redirects=True, timeout=10, headers={"User-Agent": UA})
40
        m = BVID_RE.search(str(resp.url))
41
        if m:
42
            return m.group(0)
43
    raise SystemExit(f"无法解析 BVID: {text}")
44

45

46
def load_credential() -> Credential:
47
    if not CONFIG_PATH.exists():
48
        raise SystemExit(
49
            f"找不到 credential 配置: {CONFIG_PATH}\n"
50
            "请在该路径放一个 config.toml, 内含 [credential] 段 "
51
            "(sessdata / bili_jct / buvid3 / buvid4 / dedeuserid)。"
52
        )
53
    data = tomllib.loads(CONFIG_PATH.read_text(encoding="utf-8"))
54
    cred = data.get("credential", {})
55
    if not isinstance(cred, dict) or not cred.get("sessdata"):
56
        raise SystemExit(f"{CONFIG_PATH} 的 [credential] 段缺少 sessdata。")
57
    return Credential(
58
        sessdata=cred.get("sessdata"),
59
        bili_jct=cred.get("bili_jct"),
60
        buvid3=cred.get("buvid3"),
61
        buvid4=cred.get("buvid4"),
62
        dedeuserid=cred.get("dedeuserid"),
63
        ac_time_value=cred.get("ac_time_value"),
64
    )
65

66

67
def choose_subtitle(
68
    subtitles: list[dict], preferred_lang: str
69
) -> tuple[str | None, str | None]:
70
    """优先 preferred_lang, 其次非 AI 字幕, 最后任取一个。"""
71
    for s in subtitles:
72
        if s.get("lan") == preferred_lang:
73
            return s.get("subtitle_url"), s.get("lan")
74
    for s in subtitles:
75
        if s.get("ai_type", 0) == 0:
76
            return s.get("subtitle_url"), s.get("lan")
77
    if subtitles:
78
        return subtitles[0].get("subtitle_url"), subtitles[0].get("lan")
79
    return None, None
80

81

82
def normalize_url(url: str | None) -> str | None:
83
    if not url:
84
        return None
85
    if url.startswith("//"):
86
        return "https:" + url
87
    if url.startswith(("http://", "https://")):
88
        return url
89
    return None
90

91

92
def _timestamp(seconds: float) -> str:
93
    h = int(seconds // 3600)
94
    m = int((seconds % 3600) // 60)
95
    sec = int(seconds % 60)
96
    ms = int((seconds - int(seconds)) * 1000)
97
    return f"{h:02}:{m:02}:{sec:02}.{ms:03}"
98

99

100
def format_body(body: list[dict], fmt: str) -> str:
101
    if fmt == "timestamped":
102
        chunks = []
103
        for item in body:
104
            start = float(item.get("from", 0.0))
105
            end = float(item.get("to", 0.0))
106
            content = str(item.get("content", ""))
107
            chunks.append(f"{_timestamp(start)} --> {_timestamp(end)}\n{content}")
108
        return "\n\n".join(chunks).strip()
109
    return "\n".join(str(item.get("content", "")) for item in body).strip()
110

111

112
async def fetch_subtitle(bvid: str, output_format: str) -> str:
113
    """返回字幕正文; 空串表示无字幕。"""
114
    cred = load_credential()
115
    v = video.Video(bvid=bvid, credential=cred)
116

117
    info = await v.get_info()
118
    cid = info.get("cid")
119
    if not cid:
120
        raise RuntimeError("无法确定视频 CID。")
121
    print(f"[info] cid={cid}, aid={info.get('aid')}", file=sys.stderr)
122

123
    subtitle_info = await v.get_subtitle(cid=cid)
124
    subtitles = (
125
        subtitle_info.get("subtitles", [])
126
        if isinstance(subtitle_info, dict)
127
        else []
128
    )
129
    langs = [s.get("lan") for s in subtitles]
130
    print(f"[info] 可用字幕 {len(subtitles)} 条: {langs}", file=sys.stderr)
131

132
    if not subtitles:
133
        return ""
134

135
    url, lang = choose_subtitle(subtitles, DEFAULT_LANG)
136
    url = normalize_url(url)
137
    if not url:
138
        return ""
139

140
    print(f"[info] 拉取字幕: {lang} -> {url}", file=sys.stderr)
141
    headers = {
142
        "User-Agent": UA,
143
        "Referer": f"https://www.bilibili.com/video/{bvid}/",
144
    }
145
    async with httpx.AsyncClient(headers=headers, timeout=15) as client:
146
        data = (await client.get(url)).json()
147
    body = data.get("body", [])
148
    if not body:
149
        return ""
150
    return format_body(body, output_format)
151

152

153
def main() -> None:
154
    p = argparse.ArgumentParser(description="B站视频 → 官方字幕拉取")
155
    p.add_argument("input", help="BVID / 视频URL / b23.tv 短链")
156
    p.add_argument("-o", "--output", default=None,
157
                   help="输出路径 (默认 /tmp/bilibili_subtitle_<BV>.txt)")
158
    p.add_argument("--format", default="text", choices=["text", "timestamped"],
159
                   help="字幕输出格式 (默认 text)")
160
    args = p.parse_args()
161

162
    bvid = resolve_bvid(args.input)
163
    out = Path(args.output or f"/tmp/bilibili_subtitle_{bvid}.txt")
164

165
    print(f"BVID: {bvid}", file=sys.stderr)
166
    print("拉取字幕中...", file=sys.stderr)
167

168
    try:
169
        text = asyncio.run(fetch_subtitle(bvid, args.format))
170
    except Exception as exc:
171
        raise SystemExit(f"拉取字幕失败: {exc}")
172

173
    if not text:
174
        print("无官方字幕 (若怀疑 cookie 失效, 检查 config.toml)", file=sys.stderr)
175
        print(NO_SUBTITLE_SENTINEL)  # stdout
176
        return
177

178
    out.parent.mkdir(parents=True, exist_ok=True)
179
    out.write_text(text, encoding="utf-8")
180
    print(f"字幕: {out} ({len(text)} 字符)", file=sys.stderr)
181
    print(str(out))  # stdout 只输出纯路径
182

183

184
if __name__ == "__main__":
185
    main()

1
#!/usr/bin/env python3
2
"""从B站视频下载音频 (m4a)。
3

4
无字幕时的备选方案: 下载音频交给下游转录。
5
"""
6
from __future__ import annotations
7

8
import argparse
9
import asyncio
10
import re
11
import sys
12
import tomllib
13
from io import BytesIO
14
from pathlib import Path
15

16
import httpx
17
from bilibili_api import HEADERS, Credential, get_client, video
18

19
BVID_RE = re.compile(r"BV[1-9A-HJ-NP-Za-km-z]{10}")
20
SHORT_LINK_RE = re.compile(r"^https?://(?:www\.)?(?:b23\.tv|bili2233\.cn)/")
21
CONFIG_PATH = Path.home() / ".config" / "bilibili-subtitle-fetch" / "config.toml"
22
UA = (
23
    "Mozilla/5.0 (Windows NT 10.0; Win64; x64) "
24
    "AppleWebKit/537.36 (KHTML, like Gecko) Chrome/137.0.0.0 Safari/537.36"
25
)
26

27

28
def resolve_bvid(text: str) -> str:
29
    """BVID / 视频 URL / b23.tv 短链 → BV号。"""
30
    text = text.strip()
31
    m = BVID_RE.search(text)
32
    if m:
33
        return m.group(0)
34
    if SHORT_LINK_RE.match(text):
35
        resp = httpx.get(text, follow_redirects=True, timeout=10, headers={"User-Agent": UA})
36
        m = BVID_RE.search(str(resp.url))
37
        if m:
38
            return m.group(0)
39
    raise SystemExit(f"无法解析 BVID: {text}")
40

41

42
def load_credential() -> Credential:
43
    if not CONFIG_PATH.exists():
44
        raise SystemExit(
45
            f"找不到 credential 配置: {CONFIG_PATH}\n"
46
            "请在该路径放一个 config.toml, 内含 [credential] 段。"
47
        )
48
    data = tomllib.loads(CONFIG_PATH.read_text(encoding="utf-8"))
49
    cred = data.get("credential", {})
50
    if not isinstance(cred, dict) or not cred.get("sessdata"):
51
        raise SystemExit(f"{CONFIG_PATH} 的 [credential] 段缺少 sessdata。")
52
    return Credential(
53
        sessdata=cred.get("sessdata"),
54
        bili_jct=cred.get("bili_jct"),
55
        buvid3=cred.get("buvid3"),
56
        buvid4=cred.get("buvid4"),
57
        dedeuserid=cred.get("dedeuserid"),
58
        ac_time_value=cred.get("ac_time_value"),
59
    )
60

61

62
async def _download(url: str, file) -> None:
63
    """bilibili_api 封装的流式下载 (自带绕反爬的 header)。"""
64
    download_id = await get_client().download_create(url, HEADERS)
65
    downloaded = 0
66
    total = get_client().download_content_length(download_id)
67
    while downloaded < total:
68
        chunk = await get_client().download_chunk(download_id)
69
        downloaded += file.write(chunk)
70

71

72
def _pick_url(streams, is_muxed: bool):
73
    if not streams:
74
        return None
75
    if is_muxed:
76
        return getattr(streams[0], "url", None)
77
    for stream in streams:
78
        if stream is None:
79
            continue
80
        if hasattr(stream, "audio_quality"):
81
            return getattr(stream, "url", None)
82
    return None
83

84

85
async def download_audio(bvid: str) -> BytesIO:
86
    cred = load_credential()
87
    v = video.Video(bvid=bvid, credential=cred)
88
    download_url_data = await v.get_download_url(0)
89
    detecter = video.VideoDownloadURLDataDetecter(data=download_url_data)
90
    streams = detecter.detect_best_streams()
91
    media_url = _pick_url(streams, detecter.check_flv_mp4_stream())
92
    if not media_url:
93
        raise RuntimeError("找不到可下载的音频/媒体流。")
94
    print(f"[info] 媒体流: {media_url[:80]}...", file=sys.stderr)
95
    file = BytesIO()
96
    await _download(media_url, file)
97
    file.seek(0)
98
    return file
99

100

101
def main() -> None:
102
    p = argparse.ArgumentParser(description="B站视频 → 音频下载")
103
    p.add_argument("input", help="BVID / 视频URL / b23.tv 短链")
104
    p.add_argument("-o", "--output", default=None,
105
                   help="输出路径 (默认 /tmp/bilibili_<BV>.m4a)")
106
    args = p.parse_args()
107

108
    bvid = resolve_bvid(args.input)
109
    out = Path(args.output or f"/tmp/bilibili_{bvid}.m4a")
110

111
    print(f"BVID: {bvid}", file=sys.stderr)
112
    print("下载音频中...", file=sys.stderr)
113

114
    try:
115
        audio = asyncio.run(download_audio(bvid))
116
    except Exception as exc:
117
        raise SystemExit(f"下载音频失败: {exc}")
118

119
    out.parent.mkdir(parents=True, exist_ok=True)
120
    out.write_bytes(audio.read())
121
    size_mb = out.stat().st_size / 1024 / 1024
122
    print(f"音频: {out} ({size_mb:.1f} MB)", file=sys.stderr)
123
    print(str(out))  # stdout 只输出纯路径
124

125

126
if __name__ == "__main__":
127
    main()

1
cd ~/bilibili-tools/bilibili-fetch
2

3
# 拉字幕（脚本会自动判断有无）
4
.venv/bin/python scripts/fetch_subtitle.py "https://www.bilibili.com/video/BV1xxxxxxxxx"
5
# stdout = 文件路径  → 有字幕，拿到了
6
# stdout = NO_SUBTITLE → 没字幕，改走下载音频
7

8
# 下载音频
9
.venv/bin/python scripts/download_audio.py "BV1xxxxxxxxx"
10
# stdout = /tmp/bilibili_<BV>.m4a

1
C:\Users\<你>\pi-asr\
2
├── sensevoice\                      # SenseVoice 引擎（手动放）
3
│   ├── sensevoice-small-q8.gguf     # 量化模型
4
│   ├── fsmn-vad.gguf                # VAD（人声检测）模型
5
│   └── llama-funasr-sensevoice.exe  # C++ 推理二进制
6
├── scripts\
7
│   └── transcribe.py                # 转录脚本
8
├── workspace\                       # 接收 Mac 传来的音频（自动建）
9
└── output\                          # 转录结果（自动建）

1
#!/usr/bin/env python3
2
"""通用音频转录 (Windows 端, SenseVoice / Moonshine 双引擎)。
3

4
Usage:
5
    python transcribe.py <audio_path> [-o output.txt] [--engine sensevoice|moonshine]
6
"""
7
from __future__ import annotations
8

9
import argparse
10
import subprocess
11
import sys
12
import time
13
from pathlib import Path
14

15
SENSEVOICE_DIR = Path.home() / "pi-asr" / "sensevoice"
16

17

18
# ─────────────────────────────────────────────────────────────────────
19
#  SenseVoiceSmall (C++ gguf runtime)
20
# ─────────────────────────────────────────────────────────────────────
21
def transcribe_sensevoice(audio_path: str, out_path: str) -> None:
22
    model = SENSEVOICE_DIR / "sensevoice-small-q8.gguf"
23
    vad = SENSEVOICE_DIR / "fsmn-vad.gguf"
24
    binary = SENSEVOICE_DIR / "llama-funasr-sensevoice.exe"
25
    wav_path = audio_path.rsplit(".", 1)[0] + "_16k.wav"
26

27
    for f in (model, vad, binary):
28
        if not f.exists():
29
            raise SystemExit(f"缺少引擎文件: {f}")
30

31
    # Step 1: 任意格式 → 16kHz mono WAV
32
    print("转码 (→ WAV 16kHz)...", flush=True)
33
    subprocess.run(
34
        ["ffmpeg", "-y", "-i", audio_path, "-ar", "16000", "-ac", "1", wav_path],
35
        check=True, capture_output=True,
36
    )
37

38
    # Step 2: SenseVoice 推理
39
    print("转录中 (SenseVoiceSmall, ~20x 实时率)...", flush=True)
40
    result = subprocess.run(
41
        [str(binary), "-m", str(model), "--vad", str(vad), "-a", wav_path],
42
        capture_output=True, text=True, encoding="utf-8", errors="replace",
43
    )
44

45
    output = (result.stdout or "").strip()
46
    if result.returncode != 0:
47
        output = f"[stderr]\n{result.stderr}\n\n[stdout]\n{output}"
48

49
    Path(out_path).write_text(output, encoding="utf-8")
50
    print(f"转录完成 → {out_path}", flush=True)
51
    print(f"总字数: {len(output)}", flush=True)
52

53
    Path(wav_path).unlink(missing_ok=True)
54

55

56
# ─────────────────────────────────────────────────────────────────────
57
#  Moonshine (PyTorch transformers)
58
# ─────────────────────────────────────────────────────────────────────
59
def transcribe_moonshine(audio_path: str, out_path: str) -> None:
60
    import numpy as np
61
    import torch
62
    from transformers import AutoProcessor, MoonshineForConditionalGeneration
63

64
    torch.set_num_threads(8)
65
    torch.set_num_interop_threads(8)
66

67
    print("加载 Moonshine tiny zh (PyTorch)...", flush=True)
68
    model = MoonshineForConditionalGeneration.from_pretrained(
69
        "UsefulSensors/moonshine-tiny-zh"
70
    ).to("cpu")
71
    try:
72
        model = torch.compile(model, mode="reduce-overhead")
73
        print("  (torch.compile 已启用)")
74
    except Exception:
75
        pass
76
    processor = AutoProcessor.from_pretrained("UsefulSensors/moonshine-tiny-zh")
77

78
    # ffmpeg 解码为原始 PCM (无临时文件, 支持任意格式)
79
    print("加载音频 (ffmpeg 解码)...", flush=True)
80
    proc = subprocess.Popen(
81
        ["ffmpeg", "-i", audio_path, "-f", "s16le", "-acodec", "pcm_s16le",
82
         "-ar", "16000", "-ac", "1", "pipe:1"],
83
        stdout=subprocess.PIPE, stderr=subprocess.DEVNULL,
84
    )
85
    raw = proc.stdout.read()
86
    proc.wait()
87
    audio = np.frombuffer(raw, dtype=np.int16).astype(np.float32) / 32768.0
88
    sr = 16000
89
    print(f"时长: {len(audio) / sr / 60:.1f} 分钟", flush=True)
90

91
    # 滑窗: 15s 片段, 7.5s 步长 (Moonshine max 194 tokens)
92
    chunk_n = 15 * sr
93
    stride_n = chunk_n // 2
94
    token_factor = 13 / 16_000
95

96
    results = []
97
    total = max(1, len(audio) // stride_n)
98
    t0 = time.time()
99
    print(f"共 {total} 个片段, 开始转录...", flush=True)
100

101
    for i, start in enumerate(range(0, len(audio), stride_n)):
102
        seg = audio[start:start + chunk_n]
103
        if len(seg) < sr:
104
            continue
105
        inputs = processor(seg, return_tensors="pt", sampling_rate=sr)
106
        inputs = inputs.to("cpu", torch.float32)
107
        max_len = min(194, max(1, int(
108
            (inputs.attention_mask.sum(dim=-1) * token_factor).max().item())))
109
        ids = model.generate(**inputs, max_length=max_len)
110
        text = processor.decode(ids[0], skip_special_tokens=True)
111
        if text.strip():
112
            results.append(text.strip())
113

114
        if (i + 1) % 20 == 0 or i == 0:
115
            elapsed = (time.time() - t0) / 60
116
            eta = elapsed / max(1, i + 1) * total - elapsed
117
            print(f"  {min(100, (i + 1) / total * 100):.0f}% "
118
                  f"({i + 1}/{total}) | {elapsed:.1f}分 | 预计剩余 {eta:.0f}分",
119
                  flush=True)
120

121
    full = "\n".join(results)
122
    Path(out_path).write_text(full, encoding="utf-8")
123
    print(f"\n转录完成 → {out_path}", flush=True)
124
    print(f"总字数: {len(full)}", flush=True)
125

126

127
# ─────────────────────────────────────────────────────────────────────
128
#  Main
129
# ─────────────────────────────────────────────────────────────────────
130
def main() -> None:
131
    p = argparse.ArgumentParser(description="通用音频转录 (SenseVoice/Moonshine)")
132
    p.add_argument("audio", help="音频文件路径")
133
    p.add_argument("-o", "--output", default=None,
134
                   help="输出 txt 路径 (默认与音频同目录同名 .txt)")
135
    p.add_argument("--engine", choices=["sensevoice", "moonshine"],
136
                   default="sensevoice", help="转录引擎 (默认 sensevoice)")
137
    args = p.parse_args()
138

139
    if not Path(args.audio).exists():
140
        raise SystemExit(f"音频文件不存在: {args.audio}")
141

142
    out = args.output or str(Path(args.audio).with_suffix(".txt"))
143
    Path(out).parent.mkdir(parents=True, exist_ok=True)
144
    fn = transcribe_sensevoice if args.engine == "sensevoice" else transcribe_moonshine
145
    print(f"=== 引擎: {args.engine} ===", flush=True)
146
    fn(args.audio, out)
147
    print(out)  # stdout 末行: 纯输出路径, 方便调用方捕获
148
    print("\n✅ 完成!", flush=True)
149

150

151
if __name__ == "__main__":
152
    main()

1
#!/usr/bin/env bash
2
# 通用音频转录编排 (Mac 端)
3
# 把音频 scp 到 Windows (Tailscale), 转录, 拉回纯文本。
4
# stdout 只输出文本路径, 日志全走 stderr。
5
#
6
# Usage:
7
#   transcribe_remote.sh <音频路径> [--engine sensevoice|moonshine] [-o out.txt]
8
set -euo pipefail
9

10
# ⚠️ 改成你自己的 Windows 用户名 @ Tailscale IP
11
WIN='你的用户名@你的Tailscale-IP'
12
WIN_ROOT='C:\Users\你的用户名\pi-asr'
13
WORKSPACE="${WIN_ROOT}\\workspace"
14
OUTPUT="${WIN_ROOT}\\output"
15
SCRIPT="${WIN_ROOT}\\scripts\\transcribe.py"
16

17
engine="sensevoice"
18
output=""
19
audio=""
20

21
usage() {
22
  cat <<'EOF'
23
用法: transcribe_remote.sh <音频路径> [--engine sensevoice|moonshine] [-o out.txt]
24
  默认引擎 sensevoice (中文最优), 输出 /tmp/<basename>.txt
25
EOF
26
}
27

28
while [[ $# -gt 0 ]]; do
29
  case "$1" in
30
    --engine) engine="$2"; shift 2 ;;
31
    -o|--output) output="$2"; shift 2 ;;
32
    -h|--help) usage; exit 0 ;;
33
    -*) echo "未知选项: $1" >&2; usage >&2; exit 2 ;;
34
    *) audio="$1"; shift ;;
35
  esac
36
done
37

38
if [[ -z "$audio" ]]; then
39
  echo "错误: 缺少音频路径" >&2; usage >&2; exit 2
40
fi
41
if [[ ! -f "$audio" ]]; then
42
  echo "错误: 音频文件不存在: $audio" >&2; exit 2
43
fi
44

45
basename=$(basename "$audio")
46
win_audio="${WORKSPACE}\\${basename}"
47
win_out="${OUTPUT}\\${basename}.txt"
48
# scp 路径用正斜杠（OpenSSH 在 Windows 上的约定）
49
scp_audio="C:/Users/你的用户名/pi-asr/workspace/${basename}"
50
scp_out="C:/Users/你的用户名/pi-asr/output/${basename}.txt"
51
local_out="${output:-/tmp/${basename}.txt}"
52

53
echo "[1/4] 确保 Windows 目录" >&2
54
ssh -o ConnectTimeout=15 "$WIN" "if not exist \"$WORKSPACE\" mkdir \"$WORKSPACE\" & if not exist \"$OUTPUT\" mkdir \"$OUTPUT\"" >&2
55

56
echo "[2/4] 传音频: $audio" >&2
57
scp -o ConnectTimeout=15 "$audio" "$WIN:$scp_audio" >&2
58

59
echo "[3/4] 转录中 (引擎: $engine)…" >&2
60
ssh -o ConnectTimeout=15 "$WIN" "chcp 65001 >nul && set PYTHONIOENCODING=utf-8 && python $SCRIPT \"$win_audio\" -o \"$win_out\" --engine $engine" >&2
61

62
echo "[4/4] 拉回结果" >&2
63
mkdir -p "$(dirname "$local_out")"
64
scp -o ConnectTimeout=15 "$WIN:$scp_out" "$local_out" >&2
65

66
echo "转录完成: $local_out ($(wc -m < "$local_out" | tr -d ' ') 字符)" >&2
67
echo "$local_out"

1
# Mac 上，拿到音频后
2
bash transcribe_remote.sh /tmp/bilibili_<BV>.m4a                       # 默认 SenseVoice（中文）
3
bash transcribe_remote.sh /tmp/bilibili_<BV>.m4a --engine moonshine    # 英文换 Moonshine
4
# stdout = /tmp/bilibili_<BV>.m4a.txt