README file from
GithubLexVoice
Record meetings in Obsidian, follow a live outline, and get organized notes and a mind-map overview, using your own transcription and AI services, with recordings and notes saved in your vault.
English | 简体中文 · Release notes
%%{init: {"themeVariables":{"fontSize":"14px"},"flowchart":{"nodeSpacing":8,"rankSpacing":24,"padding":8,"useMaxWidth":true,"wrappingWidth":150}}}%%
flowchart LR
root("Release review"):::lvRoot
t0("Goal"):::lvTopic0
root --> t0
d0_0["Ship this quarter"]:::lvItem
t0 --> d0_0
t1("Rollout"):::lvTopic1
root --> t1
d1_0["East region first"]:::lvItem
t1 --> d1_0
d1_1["Rollback if"]:::lvItem
t1 --> d1_1
d1_1_0["Errors rise"]:::lvItem
d1_1 --> d1_1_0
d1_1_1["Complaints spike"]:::lvItem
d1_1 --> d1_1_1
d1_2["Review Friday"]:::lvItem
t1 --> d1_2
d1_3["On-call rota"]:::lvItem
t1 --> d1_3
t2("Budget"):::lvTopic2
root --> t2
t5("Open items"):::lvTopic5
t5 <--> root
d5_0["Delay or not"]:::lvItem
d5_0 <--> t5
d5_1["Who decides"]:::lvItem
d5_1 <--> t5
d5_2["Acceptance"]:::lvItem
d5_2 <--> t5
t4("Risks"):::lvTopic4
t4 <--> root
d4_0["API change"]:::lvItem
d4_0 <--> t4
d4_1["Vendor delay"]:::lvItem
d4_1 <--> t4
t3("Owners"):::lvTopic3
t3 <--> root
d3_0["Product"]:::lvItem
d3_0 <--> t3
d3_0_0["Update spec"]:::lvItem
d3_0_0 <--> d3_0
d3_1["Engineering"]:::lvItem
d3_1 <--> t3
d3_1_0["Fix blockers"]:::lvItem
d3_1_0 <--> d3_1
d3_1_1["Load test"]:::lvItem
d3_1_1 <--> d3_1
classDef lvRoot fill:#8a8f98,fill-opacity:0.16,stroke:#8a8f98,stroke-width:1px,color:#6e7781
classDef lvTopic0 fill:#7d98cf,fill-opacity:0.18,stroke:none,color:#6e7781
classDef lvTopic1 fill:#5fb3a1,fill-opacity:0.18,stroke:none,color:#6e7781
classDef lvTopic2 fill:#e8a07a,fill-opacity:0.18,stroke:none,color:#6e7781
classDef lvTopic3 fill:#a48fd6,fill-opacity:0.18,stroke:none,color:#6e7781
classDef lvTopic4 fill:#dc8fae,fill-opacity:0.18,stroke:none,color:#6e7781
classDef lvTopic5 fill:#d9b44a,fill-opacity:0.18,stroke:none,color:#6e7781
classDef lvItem fill:none,stroke:none,color:#6e7781
linkStyle default interpolate bumpX stroke-width:1.25px
linkStyle 0,1 stroke:#7d98cf,stroke-width:1.25px
linkStyle 2,3,4,5,6,7,8 stroke:#5fb3a1,stroke-width:1.25px
linkStyle 9 stroke:#e8a07a,stroke-width:1.25px
linkStyle 17,18,19,20,21,22 stroke:#a48fd6,stroke-width:1.25px,marker-end:none
linkStyle 14,15,16 stroke:#dc8fae,stroke-width:1.25px,marker-end:none
linkStyle 10,11,12,13 stroke:#d9b44a,stroke-width:1.25px,marker-end:none
A made-up example of the meeting overview.
What it does
- Record: the microphone, system audio or both on desktop, with no virtual audio device needed on Windows; the microphone on mobile.
- Transcribe: live, after the meeting, or both. Speaker separation works where the service supports it.
- Follow along: a live outline updates as you record. During the meeting, jot notes, ask the AI a question, capture a todo, or add images and files.
- Organize: notes are generated from a template and named automatically, with a mind-map overview. Six templates are built in: General notes, Work meeting, Seminar, Interview, Personal notes and Learning notes. You can also write your own rules.
- Follow up: use Ask to check details against the notes and the raw transcript.
- Import: turn audio, MP4 screen recordings and text transcripts into notes, or watch a folder and import automatically.
- Reuse: people, todos, learning cards and hotwords become Markdown; notes about the same subject are gathered into topic pages.
- Share: generate an HTML report, a PDF report or an email draft.
- Recover: long recordings are processed in parts, and only unfinished parts are retried.
See the user manual for how each feature works.
Install
From the Obsidian Community plugins directory: open Settings → Community plugins → Browse, search for LexVoice, then install and enable it.
Manual install:
- Download
main.js,manifest.jsonandstyles.cssfrom the same official GitHub Release. - Copy them into
<your vault>/.obsidian/plugins/lexvoice/. - Reload Obsidian and enable LexVoice under Community plugins.
When updating manually, replace only these three files and keep data.json and your vault files.
First recording
- Open Home in LexVoice settings. Under Quick setup, choose a service, paste an API key and choose Test and enable. If you have no account, choose Sign up.
- Open the LexVoice sidebar and choose a Template and Audio input.
- Choose New recording and check that the level meter reacts. Fill in the name, speaker and outline in the note's Meeting info box, or leave it empty.
- Watch the live outline and add in-meeting notes if needed.
- Stop recording, then click the LexVoice status in the status bar to open the task center and follow transcription and AI organization. If processing was interrupted, continue from where it stopped.
- Open Knowledge, choose Scan this note, and review people, todos, learning cards and hotwords.
- To share, right-click a note in the Notes list, choose Generate, then an HTML report, PDF report or email draft.
Recordings go to LexVoice/录音 and notes to LexVoice/转写纪要 by default. Both can be changed in settings.
What's new in 3.1.3
- Built-in meeting notes can bring together remarks that answer the same question, even when they occur at different points in the discussion.
- Notes may use a table when comparing items side by side makes a point clearer and the recording supports the comparison. They can use lists for parallel points and paragraphs for reasoning, context and exceptions. The prompts discourage filling gaps with guesses.
Requirements
- Obsidian 1.10.0 or later.
- A speech-to-text service, cloud API or local.
- An AI service is recommended: the live outline, organized notes, the overview, knowledge extraction and reports all use it.
- A virtual audio device to record system audio on macOS or Linux.
| Use | Services |
|---|---|
| Transcription and AI | SiliconFlow, Alibaba Cloud Bailian, OpenAI, Xiaomi MiMo |
| AI only | DeepSeek, Moonshot / Kimi, Zhipu GLM, Volcengine Ark, Tencent Hunyuan, MiniMax, Baidu Qianfan, OpenRouter, Google Gemini and others |
| Local | A local transcription service, Ollama, LM Studio, a local OpenAI-compatible service |
| Custom | Any OpenAI-compatible AI service |
To use different services for different tasks, add them under Model services, then choose them in Transcription & AI.
The interface follows Obsidian's language and is available in Chinese, English and Japanese.
Privacy
No ads, no analytics, no telemetry, and no LexVoice server. Settings are stored in .obsidian/plugins/lexvoice/data.json; recordings and notes stay in your vault.
With a cloud service, the relevant audio, transcript text and prompt context are sent to the provider you chose. For sensitive content, prefer local transcription and a local model, and obtain consent before recording.
- Anything in the Meeting info box is always sent to the configured AI service, and to the transcription service if it accepts prompts. Leave it empty to send nothing from it.
- Preparation material is sent only after Use for transcription & AI is on. Turning it off does not recall requests already sent.
- Handwritten notes are not sent to AI when generating notes, overviews or reports; knowledge scans and AI polish on a whole note do read them.
- Automatic links and meeting topics is on by default. Background analysis may incur provider charges; it does not additionally send audio, paths or full verbatim transcripts. Turning it off does not undo content already written.
See PRIVACY.md for details.
FAQ
Yes. LexVoice ships no keys and has no account.
Yes. Mobile recording uses the device microphone. Keep Obsidian in the foreground, because locking the screen or switching apps may interrupt recording. System audio, virtual audio devices, multichannel capture, device tests, streaming live transcription and PDF reports require desktop.
Windows: under Audio, choose Microphone + Windows system audio (recommended), or Windows system audio only for video and courses. If the direct test fails, use VB-Cable.
macOS: use BlackHole. Linux: use a PulseAudio / PipeWire monitor source.
Before a long recording, run Test devices in Settings → Advanced → Audio input and play some audio during the test to confirm sound.
Audio in MP3, WAV, M4A, AAC, WebM, Ogg, FLAC and similar formats, plus Markdown and TXT. On desktop, MP4 video uses only its audio track. WebM video and most files over 1 GB are refused with an explanation.
Several providers have these built in. Open a connection in Model services to switch billing. Most plans offer AI organization only, not transcription. The terms of many plans prohibit use from apps like LexVoice, so check yours first.
Run Copy diagnostic report and send it to the developer. The report has no audio, transcript text or full prompts, and common API keys, user folders and vault paths are hidden automatically. See the user manual for troubleshooting.
License & credits
Beginning with the 2.2.0 release line, new LexVoice material is distributed under the LexVoice Proprietary Software License. Development source code is no longer publicly released. Official JavaScript runtime files remain inspectable; that does not make this an open-source release or grant a right to redistribute the runtime or source code.
Personal and internal business use is permitted. Your recordings, notes, and exported reports may still be edited, shared, and commercially used, subject to rights in their contents. Unauthorized rebranding, repackaging, external redistribution (free or paid), resale, and white-label software offerings are prohibited. Local adjustments for your own permitted use are allowed; distributing a modified plugin is not.
Previously MIT-licensed releases, including 2.1.2, and previously MIT-licensed portions reused in later releases retain their original rights. See the preserved MIT notice. Third-party components retain their own licenses; see Third-Party Notices. Nothing here restricts mandatory legal rights or independent implementations of general ideas.
Closed-source distribution through the Obsidian community directory is subject to Obsidian's case-by-case review. This license change does not itself establish approval for the new distribution model.
The HTML slide-deck feature in earlier versions was inspired by alchaincyf/huashu-design; its HTML-first slide workflow and design principles influenced that work. Per the upstream license: Derived from alchaincyf/huashu-design.