功能詳解All features

每一項功能，完整說清楚Every feature, in full

從採用的 AI 模型、10 種功能模式，到關鍵字通知、懸浮字幕與字幕轉發——這裡是 jt-live-whisper 的完整功能地圖。From the AI models and 10 modes to keyword alerts, overlay subtitles and forwarding — the complete feature map.

使用的 AI 模型AI models used

所有模型皆在自有設備上推論（本機或區網內的 GPU 伺服器），不需要任何第三方雲端 API。Every model runs on your own hardware (local or a LAN GPU server) — no third-party cloud API.

用途Purpose AI 模型Model 說明Notes
語音辨識 (ASR)ASRwhisper.cppmacOS 即時辨識引擎，支援中日英文，可在本機或 GPU 伺服器執行macOS live engine; ZH/JA/EN; local or GPU server
語音辨識 (ASR)ASRfaster-whisper (CTranslate2)Windows 即時辨識 + 全平台離線處理，支援 VAD 靜音過濾Windows live + all-platform offline; VAD silence filtering
語音辨識 (ASR)ASRmlx-whisperApple Silicon GPU 加速，雙向模式（en_zh / ja_zh）即時辨識專用Apple Silicon GPU; for bidirectional live modes
語音辨識 (ASR)ASRMoonshine (Useful Sensors)超低延遲串流辨識模型，英文專用（僅限 Apple Silicon）Ultra-low-latency streaming, English only (Apple Silicon)
翻譯 / 摘要Translate / summary自架 LLM 伺服器（推薦 Qwen / Phi-4 / GPT-OSS）Self-hosted LLM (Qwen / Phi-4 / GPT-OSS)透過地端 Ollama 或其他 LLM 伺服器，翻譯建議 14B 以上、摘要建議 120B 以上Via local Ollama or other servers; ≥14B for translation, ≥120B for summaries
翻譯 (離線)Translate (offline)NLLB 600M (Meta)離線翻譯，支援中日英互譯（en2zh / zh2en / ja2zh / zh2ja）Offline ZH/JA/EN translation
翻譯 (離線備援)Translate (fallback)Argos Translate完全離線的輕量翻譯模型，僅支援英翻中Fully offline, lightweight; English→Chinese only
講者辨識Diarizationresemblyzer + spectralcluster聲紋特徵提取 + 頻譜分群，可在本機或 GPU 伺服器執行Voice embeddings + spectral clustering; local or GPU server

為什麼講者辨識不用 pyannote.audio？ pyannote 的預訓練模型授權限制了用途與場景，且需要在 HuggingFace 註冊帳號、申請存取權限並設定 Token 才能下載。這不符合本工具「零帳號、零註冊、完全地端」的設計理念。resemblyzer + spectralcluster 完全開源、安裝即用、無需任何帳號或 Token。 Why not pyannote.audio for diarization? Its pretrained models carry usage restrictions and require a HuggingFace account, access request and token. That clashes with the project's "no account, no signup, fully on-device" principle. resemblyzer + spectralcluster are fully open source and work out of the box.

10 種功能模式10 functional modes

單向翻譯、雙向翻譯、純轉錄、純錄音，滿足各種使用場景。One-way, bidirectional, transcription-only and record-only — for every scenario.

en2zh英翻中EN → ZH
zh2en中翻英ZH → EN
ja2zh日翻中JA → ZH
zh2ja中翻日ZH → JA
en_zh英中雙向EN ↔ ZH 系統音訊 + 麥克風system + mic
ja_zh日中雙向JA ↔ ZH 系統音訊 + 麥克風system + mic
en純英文轉錄English transcription
zh純中文轉錄Chinese transcription
ja純日文轉錄Japanese transcription
record純錄音Record only

所有即時轉錄模式加上 --mic 即可同時轉錄自己的麥克風語音，雙向模式則自動啟用雙路辨識。Add --mic to any live transcription mode to also transcribe your microphone; bidirectional modes enable dual-stream automatically.

多種本地端引擎Multiple local engines

辨識與翻譯都有多種引擎可選，依場景與硬體自由搭配。Choose recognition and translation engines to fit your scenario and hardware.

語音辨識引擎Recognition engines

  • Whisper：高準確度，即時辨識主力Whisper: high accuracy, the live default
  • Moonshine：超低延遲 ~300ms（英文）Moonshine: ultra-low latency ~300 ms (English)
  • faster-whisper：離線批次處理，支援 VAD 靜音過濾faster-whisper: offline batch, with VAD silence filtering

翻譯引擎Translation engines

  • LLM：Ollama / OpenAI 相容伺服器，品質最佳LLM: Ollama / OpenAI-compatible servers, best quality
  • NLLB：離線中日英互譯，無需伺服器NLLB: offline ZH/JA/EN, no server needed
  • Argos：離線備援（僅英翻中）Argos: offline fallback (EN→ZH)

支援的本地端 LLM 伺服器Supported local LLM servers

程式會自動偵測 LLM 伺服器類型，不需手動選擇。The app auto-detects the server type — no manual selection.

伺服器Server預設 PortDefault portAPI
Ollama11434Ollama 原生Ollama native
LM Studio1234OpenAI 相容OpenAI-compatible
Jan.ai1337OpenAI 相容OpenAI-compatible
vLLM8000OpenAI 相容OpenAI-compatible
LocalAI / llama.cpp8080OpenAI 相容OpenAI-compatible
LiteLLM4000OpenAI 相容OpenAI-compatible

進階特色Advanced features

把即時翻譯變得更實用的那些細節。The details that make live translation genuinely useful.

同時轉錄麥克風Transcribe your mic

所有即時模式加上 --mic 即可同時轉錄自己的麥克風語音，雙向模式自動啟用。Add --mic to any live mode to also transcribe your own voice; auto-on for bidirectional.

會議主題感知翻譯Topic-aware translation

可指定會議主題（如「ZFS 儲存管理」），讓 LLM 依領域上下文精準翻譯專業術語。Set a meeting topic (e.g. "ZFS storage") so the LLM translates domain jargon accurately.

自動偵測 LLM 伺服器Auto-detect LLM server

支援 Ollama、LM Studio、Jan.ai、vLLM、LocalAI、llama.cpp、LiteLLM，自動辨識伺服器類型。Detects Ollama, LM Studio, Jan.ai, vLLM, LocalAI, llama.cpp and LiteLLM automatically.

互動式選單 + CLIInteractive menu + CLI

新手友善的選單介面，進階用戶可用命令列參數直接啟動；選單最後顯示等效 CLI 指令。A beginner-friendly menu, or launch directly via CLI — the menu prints the equivalent command.

WebUI 瀏覽器介面Web UI

--webui 在瀏覽器操作所有功能，支援即時字幕、離線處理、講者辨識、摘要，手機 / 平板也可使用。--webui drives everything in the browser — live, offline, diarization, summaries — phone/tablet friendly.

背景降噪Background denoise

即時模式可加 --denoise 啟用背景降噪，提升嘈雜環境的辨識品質。Add --denoise in live mode to clean up noisy environments before recognition.

關鍵字即時通知Keyword alerts

設定關鍵字，即時辨識出現時自動發出通知，可用於追蹤會議重點，或線上課程摸魚時讓系統在「請實作」「這個會考」時自動提醒。支援全螢幕警示特效、瀏覽器推播、音效提示（警示 / 柔和可選）、懸浮字幕閃爍，同一關鍵字冷卻機制避免重複通知。Set keywords and get notified the moment they're spoken — track key topics, or get pinged on "let's implement this" / "this is on the exam". Full-screen effects, browser push, sound (alert / soft), overlay flashing, and a per-keyword cooldown.

關鍵字通知效果

懸浮字幕Overlay subtitles

桌面半透明字幕覆蓋視窗（PyQt6），可疊加於任何應用程式上方。字體依視窗大小自動縮放、可拖曳移動與調整大小、滑鼠穿透模式、字幕切換淡入淡出動畫，單語 / 雙語自動切換高度。（感謝 OSSLab 熊大提供建議）A translucent desktop overlay (PyQt6) over any app — auto-scaling text, drag & resize, click-through, fade animations and auto height for mono/bilingual lines.

懸浮字幕效果

字幕轉發Subtitle forwarding

即時字幕自動轉發到通訊平台（Telegram / Slack / Discord / Teams / LINE / Nextcloud Talk / 通用 API），可同時啟用多個平台、自訂發送間隔與內容（含時間 / 原文 / 譯文）。通用 API 支援 Body 範本（{{text}} 變數）搭配自訂 Headers。Forwards live subtitles to Telegram / Slack / Discord / Teams / LINE / Nextcloud Talk / a custom API — multiple at once, with custom intervals and content. The custom API supports a body template ({{text}}) with custom headers.

Telegram 轉發效果

產出檔案Output files

離線處理的產物，都存於 logs/<session>/。Offline processing writes everything to logs/<session>/.

檔案File說明Description需要 LLMNeeds LLM
時間逐字稿_*.txt帶時間戳逐字稿（翻譯模式含原文 + 譯文）Timestamped transcript (source + translation in translate modes)校正需要for correction
時間逐字稿_*.html互動式逐字稿（點時間戳可播放音訊）Interactive transcript (click timestamps to play)校正需要for correction
時間逐字稿_*.srtSRT 字幕檔SRT subtitles否No
時間逐字稿_*.vttWebVTT 字幕檔WebVTT subtitles否No
摘要_*.txtAI 重點摘要 + 校正逐字稿AI summary + corrected transcript是Yes
摘要_*.htmlAI 摘要 HTML（含樣式與相關檔案連結）AI summary HTML (styled, with links)是Yes

有設定 LLM 伺服器時，逐字稿會自動經過 LLM 校正（修正 ASR 辨識錯字），純轉錄模式同樣支援。With an LLM server configured, transcripts are auto-corrected by the LLM (fixing ASR typos) — including transcription-only modes.

品質與效能說明Quality & performance

1語音辨識品質Recognition quality

取決於所選 ASR 模型大小、音訊品質（背景噪音、麥克風距離、多人重疊）以及語言種類。Depends on model size, audio quality (noise, mic distance, overlap) and language.

2翻譯品質Translation quality

取決於翻譯引擎與模型能力。LLM 最佳但需伺服器；NLLB / Argos 離線品質較低但無需伺服器。Depends on engine and model. LLM is best (needs a server); NLLB / Argos are offline but lower quality.

3講者辨識準確度Diarization accuracy

受音訊品質、講者數量與聲紋相似度影響，多人交談或遠場收音時可能不準。Affected by audio quality, speaker count and voice similarity; crosstalk and far-field hurt accuracy.

4處理速度Processing speed

取決於硬體算力（CPU / GPU）與模型大小。GPU 伺服器可大幅加速；純 CPU 較慢。Depends on hardware (CPU/GPU) and model size. A GPU server is much faster; CPU-only is slow.

免責聲明：本工具按「現狀」（AS IS）提供，不附帶任何明示或暗示的保證。語音辨識、翻譯、講者辨識及摘要等功能的輸出僅供參考，不保證準確性與完整性。使用者應自行驗證輸出，不應將未經人工審核的輸出直接用於法律文件、醫療紀錄、財務報告或其他需要高度準確性的場合。使用者應確保擁有合法錄音權利並遵守當地隱私法規。作者及貢獻者不對因使用本工具而產生的任何損害承擔責任。 Disclaimer: Provided "AS IS" without warranty of any kind. Recognition, translation, diarization and summary outputs are for reference only and are not guaranteed to be accurate or complete. Verify outputs yourself; do not use unreviewed output for legal, medical, financial or other high-stakes purposes. Ensure you have the legal right to record and comply with local privacy laws. The author and contributors are not liable for any damages arising from use.