The AI tools index that doesn't waste your time.不浪费你时间的 AI 工具索引。

Whisper

OpenAI's open-source speech recognition model, robust and multilingual.
OpenAI开源的多语言语音识别模型
Open source开源★ Editor's pick精选#speech-to-text#open-source#transcription
Visit site访问官网公式サイトへ ↗github.comLink active链接活跃Probed 33d ago33 天前探测 (2026-09-19)(2026-09-19) · 762 ms · 100% up · HTTP 200 ·
✎

Editorial review编辑测评

2026-10-15

The open-source workhorse for private, high-volume transcription — aging core, unbeatable economics.隐私与大批量转写的开源主力——核心渐老,但成本结构无可匹敌。

Whisper is the speech recognition model that reset the baseline: open-source, multilingual, and robust enough that 'just run Whisper' became the default answer to transcription. Its strength is robustness under real-world audio — accents, background noise, code-switching between languages — where older models fell apart. Being fully open-source (MIT) changes the economics completely: no per-minute fees, no data leaving your machine, and a model size ladder from tiny to large that lets you trade accuracy for speed on whatever hardware you have. The ecosystem is arguably the bigger asset now: faster-whisper and whisper.cpp ports make it run comfortably on a laptop CPU. Caveats from our side: the GitHub repo is reachable but cn_ok is false in our directory, so mainland users should expect occasional GitHub access friction when pulling weights. Our probe shows the repo page at 100% uptime over a month of daily checks, though response averaged around 762 ms — GitHub being GitHub. Honest weaknesses: the vanilla package shows its age — no built-in speaker diarization, timestamps drift on long files without VAD tuning, hallucinations appear on silent segments, and OpenAI has not shipped a major update in a while, leaving Chinese-language edge cases to community forks. For privacy-sensitive, high-volume, or offline transcription, it remains the workhorse to beat.

Whisper 是重新定义了基准线的语音识别模型:开源、多语言、鲁棒到“直接跑 Whisper”成了转写问题的默认答案。它最强的是真实音频下的抗造能力——口音、背景噪声、中英混说,这些老模型翻车的地方它都扛得住。完全开源(MIT)彻底改变了成本结构:没有按分钟计费,数据不出本机,从 tiny 到 large 的模型尺寸梯度让你按硬件在精度和速度之间自己权衡。生态现在可能是更大的资产:faster-whisper 和 whisper.cpp 移植让它在笔记本 CPU 上都跑得舒服。我们这边的注意事项:GitHub 仓库可达,但目录中 cn_ok 为 false,国内用户拉权重时要对 GitHub 偶发抽风有准备;探活显示仓库页一个多月每天 200,不过平均响应约 762ms,GitHub 日常水平。短板说实话:原版包已显老态——没有内置说话人分离,长音频时间戳不配 VAD 会漂,静音段会产生幻听,OpenAI 也很久没有大版本更新,中文边角场景基本靠社区 fork 补。隐私敏感、大批量、离线转写场景,它仍是要被打败的那个标杆。

Pros优点

  • Fully open-source (MIT): zero per-minute cost, audio never leaves your machine
  • Model size ladder plus faster-whisper/whisper.cpp ecosystem runs anywhere
  • Robust to accents, noise and language code-switching where older ASR breaks

Cons缺点

  • No built-in speaker diarization; long-file timestamps drift without VAD tuning
  • Core model is aging with no recent major update; GitHub access friction in China (cn_ok: false)

Best for: 最适合:Developers and teams needing private, offline, or high-volume speech-to-text without per-minute bills.需要私有化、离线或大批量语音转文字、又不想付按分钟账单的开发者与团队。

◆

Appears in scenarios出现在这些场景

1 scenario
⚖

Compared with与它对比

1 comparison
◆

Appears in workflows出现在工作流

2 workflows
?

Quick answers常见问答

Is Whisper free?Whisper 是免费的吗?

Whisper is open source — free to self-host.Whisper 开源,可自行部署免费。

Does Whisper work in China?国内能直接用 Whisper 吗?

Whisper may require a proxy from mainland China; we recheck its link daily.Whisper 国内可能需要代理;我们每日复测其链接。

Is Whisper's link verified?Whisper 的链接核验过吗?

Yes — we probed Whisper's link 33d ago (HTTP 200, 762 ms) and it's live.是——我们33 天前探测 Whisper 链接(HTTP 200, 762 ms),状态活跃。