README file from
GithubCozy Read Aloud Mobile
A mobile-first read-aloud companion for Markdown notes, text-based PDFs and local HTML files in Obsidian.
Read from a compact mini outline. Navigate chapters, start a section from Outline options, and pause/resume in the outline header without opening the main playback bar. Touch controls stay visible. See the desktop walkthrough of this workflow; it is not a phone recording, and current-release physical-device validation remains pending.
Formerly Note and PDF Voice Reader for Mobile. Same plugin, new name. Your settings and update path remain unchanged; no reinstall is required.
Quick Demos
Open the toolbar, read, pause and close (26 seconds).
Silent demos recorded in Obsidian's mobile-layout simulation on Windows, not on a phone. Recording details.
Select a passage, expand the player, choose Read selection, then use Back to document to return without stopping playback. Selection gestures differ on real phones.
This earlier demo shows the full outline dialog, still available from the Open PDF outline command. The toolbar's outline button now opens the mini outline: tap a title to navigate, or use the section actions in the top Outline options to read from it or read only that section. Navigation does not start or restart playback.
Highlights
-
0.8.5 table wording: The academic reading preview formats fixed examples locally without using notes or a speech service. Repeat table headers for each row is on by default. Turn it off to read headings once and then row values; smart omission remains independent. Release notes.
-
0.8.4 formula and control update: shared Core 0.3.0 adds optional connector-free simple functions and range-only interval speech in concise style. Optional floating 0.25x speed steps report the current rate; secondary-mouse volume steps are separately opt-in. Existing voices and defaults are preserved. Release notes.
-
Mini outline playback: choose a heading, use the top Outline options to read that section or continue from it, then pause/resume in the same panel. Tapping a title alone only navigates. Search and pin the outline without opening the main playback bar; no double-tap gesture is required.
-
Optional OCR (experimental): off by default, with a separately installed resource pack and single-page recognition. Read printed text in scanned PDFs and PNG/JPEG/WebP after editing the preview. No image upload, no model load at startup. Physical iPhone/iPad/Android validation is pending. Setup and device checklist.
-
Read beside your document: touch-sized controls with pause, Stop, speed and current-audio seeking.
-
Passage highlighting: a soft highlight marks the current passage in Markdown reading/editing views, text-based PDFs and accessible local HTML Reader views. Toggle the highlighter icon or Playback > Highlight reading passage. This is passage-level, not word timing or automatic scrolling. Uncertain matches are left unmarked; changing the source clears stale marks. Highlighting does not modify files or send extra text to speech providers.
-
Editor menu: Markdown Live Preview/Source context menus offer Read aloud from here and Read selected text. Mobile availability depends on Obsidian exposing its editor menu; the system selection popup is not replaced. Existing toolbar actions remain available for PDF, HTML and reading views.
-
Academic PDF reading: column-aware text ordering, optional header/footer filtering, and conservative footnote separation, including first-page correspondence blocks.
-
PDF formula handling: when Strip Markdown is enabled, local font evidence can repair known math-font encoding errors. Smart mode skips complex equations while retaining readable short expressions. Full-file, section and selected-text reading use the same rules; ambiguous selections ask you to select again rather than read nearby text. This is not OCR or a guarantee for every PDF font.
-
Document outlines: navigate Markdown headings, HTML h1-h6 headings and PDF bookmarks/inferred headings. Tapping a title locates the source without starting speech. The mini outline's top Outline options provides the separate section-reading actions.
-
Sentence-first chunks: move a complete sentence to the next chunk when it cannot fit; split only oversized sentences at clause/word boundaries within the provider cap. Conservatively join PDF cross-page continuations.
-
Local HTML: open files with HTML Reader, then read the whole file, a selection or from a selected position.
-
WAV export: save an online-engine reading range to the vault after explicit confirmation.
-
Privacy choices: device speech by default; optional online providers require permission. No developer relay server or built-in telemetry.
Install and Start
- Use Obsidian 1.11.4 or later and back up your vault.
- Open Settings → Community plugins → Browse and search for Cozy Read Aloud Mobile.
- Select Install, then Enable. You can also open the community listing.
- Open a note or PDF and tap the reader icon to show the playback toolbar.
- Start with device system speech, or choose an online engine in the plugin settings and explicitly enable processing after configuring its secret.
To browse and read local HTML files, install and enable HTML Reader as well. Avoid starting the desktop and mobile reader plugins at the same time in one vault.
Voice and shortcut options
Settings separate Reading method (System speech / Online speech / Self-hosted service) from the online provider. Existing configurations are preserved. Configure the secret and voice, review the processing permission, then use Test voice for a fixed sample that never reads your notes. Online tests may incur charges. Stop test cancels the test; it does not replace an ongoing reading or export. Self-hosted speech uses your HTTPS endpoint, not a model running on the phone. Advanced chunk and prefetch controls are collapsed in Playback settings.
Playback settings offer independent switches for the editor reading menu and the optional Editing Toolbar speaker icon. The icon defaults to selected text, with an option to read from the selection to the document end. It requires a selection and an enabled Editing Toolbar plugin.
Compatibility and Testing
Mini Outline
0.8.0: the mini outline has a pause/resume button independent of the main playback toolbar. It follows the header's idle hiding by default on desktop; Mini outline > More outline options > Playback button visibility can keep it visible during reading instead. Touch screens always retain it. The optional Floating playback controls setting in Reading shortcuts adds one lower-right control in the current Markdown/PDF/local HTML pane. It is off by default, appears only for an active/paused session, and never starts new speech. Both controls share the existing player state. Physical-device validation remains pending. Controls and limits.
Under Floating playback controls > More floating control options, Show stop reading button is on by default. The options start collapsed; collapsing them does not disable Stop. Stop appears above Pause on desktop hover/keyboard focus and stays visible on touch. Collapsed chapter lines use accent color alone for reading, with no dot or thickness change.
0.8.1: optional Show 5-second seek buttons adds Back/Forward beside Pause. Off by default, visible on touch when enabled, and hidden if space is insufficient. Seeking uses the current loaded online audio and is disabled for device system speech. After unpinning/collapsing, the outline entry follows the active note and remains available when returning; touch Auto still uses an icon. Markdown/local HTML refresh only the active outline; PDFs use a cached outline or retain an icon until opened. Explicit hiding and Floating TOC coexistence are preserved. Mobile background playback remains experimental; current-release physical-device testing is pending.
Floating playback demo with captions (desktop recording): hover to reveal Stop and Back/Forward, pause, seek five seconds, resume and stop. Touch screens keep the enabled extra controls visible. The clip uses desktop local speech and is not a mobile-device test; mobile device system speech does not support these seek buttons.
Open the document-edge directory button, the reading toolbar's outline icon, or Toggle mini outline without starting speech. Search headings, choose expanded levels and pin the directory. Titles use two lines by default, configurable from one to four, and navigate only; no menu buttons appear beside them. The top Outline options identifies the target section and offers full-heading expansion, reading from that section or reading only that section and its descendants. With a mouse or keyboard, the same actions are available by right-clicking a heading or pressing Shift+F10. Common Markdown formatting is cleaned only in directory labels; navigation and source text are unchanged. A shaded row marks browsing; accent text and a small play marker identify the reliably located reading section. Manual directory browsing pauses tracking until the Return to reading arrow is used. The list scrolls independently and reserves bottom space.
Adjust the width from 180 to 600 pixels, with automatic limits in narrow panes. Reading shortcuts appear above Mini outline in settings; secondary preferences are under More outline options. Touch header icons and Close stay visible, wrapping at smaller widths without shrinking their hit targets. Use Go to document top/bottom in Outline options to navigate without starting speech. On desktop, icons and scrollbar thumbs appear on hover or keyboard focus with an idle pin indicator; optional mouse double-click reading is off by default. Touch scrolling feedback is unchanged. Touch unpinning collapses immediately; with a mouse it waits until the pointer leaves.
Collapsed outline (0.8.0): More outline options > Collapsed appearance offers Auto / Chapter lines / Outline icon. Auto uses a 44-pixel icon target for touch and subtle chapter lines for a mouse. Tap to reopen on touch; mouse hovering over lines opens the outline without navigation or speech. Dense outlines are summarized, not shown as page percentages. Until headings are loaded, the entry remains an icon without extra PDF analysis. Explicit hiding and Floating TOC coexistence still take priority. Physical-device validation remains pending.
In Playback → Mini outline, the Mini outline button setting offers Auto, Always show or Hide button. Auto hides our document-edge button when Floating TOC or another recognized floating-outline plugin is enabled; toolbar and command access remain. Choose left/right placement and optional mouse hover; touch uses tap. The compact icon bar reveals search and expanded-level options on demand. Transparent background is optional on desktop; phones and tablets retain a solid background and larger controls. Reading shortcuts have their own settings group. No other plugin is modified. The directory does not scan the vault. Its state is memory-only; changed files invalidate old targets. This replaces basic floating chapter navigation, not every specialized TOC feature. Behavior and limits.
Open mini outline on startup is off by default. Enable it to open and pin the active outline after workspace restoration, hiding its edge entry while expanded and without starting speech. Only the active supported document is loaded; background tabs are not parsed. With the option off, startup does not read text or parse PDFs. Dismissing the outline prevents further automatic opening during that session; its entry then follows the Mini outline button preference. Physical-device validation remains pending.
A pinned, open outline stays visible when you switch documents and updates to the active document's headings. Closing or unpinning it stops automatic following; playback and keyboard focus remain unchanged.
Device Validation
For this release, use the step-by-step device checklist and public sample. Current-device acceptance is pending; automated interruption and selection tests are not physical-device results.
Earlier iPhone testing confirmed full-file reading, starting from selected text, pause/resume, Stop, speed adjustment and an unobstructed toolbar in the user's setup. This is feedback on an earlier release, not a full device test of each subsequent update.
Compatibility can vary by device, theme and PDF layout. Outlines, footnotes, HTML reading and audio export still need broader device testing, including on iPad and Android. Complex PDFs may have layout-specific issues. See the device checklist for the earlier reports and scenarios to verify.
Background playback is experimental and off by default. iOS may interrupt playback when the app is in the background or the screen is locked. iPhone/iPad background continuity has not yet passed physical-device testing; the option applies only to online/self-hosted audio, not system speech.
Current capabilities
Phone controls
The reader icon and Show reading toolbar command open a compact bar above the current note or PDF, below its title bar, without switching to a separate tab. The controls reserve their own space outside the document scroll area to avoid iOS floating bottom navigation. Select a reading scope and press Play. Pause/resume, previous/next chunk, current-audio seeking and playback speed are available beside the document. Seeking only applies to loaded online audio, not system speech.
The square Stop button ends reading or cancels PDF extraction while keeping the bar open. Expand opens the full player. Its back arrow returns to the document without stopping playback; Close stops and hides the reader. Select text and choose Continue from selection to start there. The source selection is retained in memory before opening the panel, not written to settings; select it again after editing the document. Stopping cannot undo a speech request already sent to a provider or its charges.
On phones, opening the reading bar temporarily keeps the current pane's native title controls in normal layout below the status-bar safe area; closing the bar restores the original layout. On iPad, the bar uses a single row in wide panes and stacked rows in narrow Split View or Stage Manager windows. It does not apply phone-specific status padding to tablet panes. iOS/Android device verification is still needed, particularly for rotation, keyboards, native selection menus and PDF views.
- Read selected note or PDF text
- Continue reading from a selected position
- Read an entire active note or text-based PDF
- Resume from a privacy-bounded saved position
- Coordinate-aware single-column, two-column, and mixed-layout PDF ordering
- Optional local PDF header/footer filtering, enabled by default
- PDF bookmarks or an inferred outline, section reading and a one-document in-memory outline cache
- Body-only, original-order or footnotes-only PDF reading
- Local HTML reading and confirmed WAV export for online speech engines
- Device system speech with no text sent by the plugin to an online TTS service
- Optional Xiaomi MiMo, Microsoft Azure Speech, OpenRouter TTS, HTTPS remote CosyVoice and a custom speech API (BYOK)
- Touch-sized controls, vertical layouts, pause/resume, and chunk navigation
- Online synthesis is on demand by default; optional one-part lookahead reduces gaps
- Local online-audio playback speed, volume and seeking within the currently loaded audio
- Categorized settings, academic text handling and progressive opening audio within logical chunks
Online Setup
Price references: OpenRouter settings show dated prices for verified ElevenLabs and MAI presets, character-based cost estimates and an official-pricing link. These are offline snapshots, not live quotes. The recorded ElevenLabs offer ends on October 19, 2026 at 15:00 UTC; settings opened afterward show only the last checked list price and a reminder to verify it. Other models show no numeric estimate. Repeat synthesis, fees and taxes can increase the final cost.
ElevenLabs via OpenRouter: choose Multilingual v2, Flash v2.5, Eleven v4 or v4 Turbo, with Sarah as the default and George, Daniel, Alice, Brian and Lily also available. Existing saved voice choices are preserved. Audio uses MP3 at normal synthesis speed; adjust playback speed in the player. ZDR and denial of data collection remain mandatory; an unavailable route fails without relaxing privacy. The speech catalog and ZDR endpoint list were checked on 2026-10-08. Prices and availability can change.
Optional continuity: these four models show an ElevenLabs advanced section with an experimental, default-off continuity toggle. It sends only neighboring sentences within the current reading range, up to 160 characters per side, through provider.options.elevenlabs; it never reads outside a selected range or uses request history. Other models hide the option and receive no context parameters, while the saved preference is retained. No cloud pronunciation dictionary or automatic performance tags are added. Source audio tags may affect delivery. Parameter support follows the OpenRouter announcement; paid listening tests have not been performed, so improved sound is not guaranteed.
Local term pronunciations: in Academic settings, enable rules such as AI = artificial intelligence, one per line. Rules are case-sensitive literal matches (not regex), longest match first, without recursive replacement. Up to 100 rules are stored locally; each term is limited to 80 characters and each pronunciation to 120. Online engines receive the substituted speech text, not the rule list. Source documents and original reading positions are preserved; expanded requests are split to respect engine limits. These rules also apply to export. Rules take effect next session; edited rules do not change already-prepared audio.
Formula style: concise reads r_t as "r t", r_{t+1} as "r t plus 1", and x_{i,j} as "x i j". Verbose style retains "subscript"; compound indices keep structural cues. Markdown, local HTML and PDF share this conversion, with smart-mode omission unchanged. Changes apply to the next reading session, not audio already synthesized.
Concise short labels include \widehat q / \hat q as "q hat", \widetilde{x} as "x tilde", \ddot{x} as "x double dot", x' / x^{\prime} as "x prime", q^{\ast} as "q star", A^{\top} as "A transpose", and \vec{x} as "vector x". Chinese formula speech uses Chinese vector/transpose labels. Indices stay ordered: \widehat{q}_t becomes "q hat t"; \bar{x}_{i,j} becomes "x bar i j". No estimate, derivative or optimum is inferred, and ordinary A^T remains a power. Verbose style and complex-formula protection are unchanged. PDF prime glyphs and superscripts need reliable text/font/position evidence; arbitrary accent layouts are not reconstructed.
Xiaomi MiMo: create a key in the MiMo console, select its Obsidian secret, and enable processing. The default voice is 冰糖 (Bing Tang, Chinese female); 白桦 (Bai Hua) remains available for Chinese and English, and existing voice choices are preserved. MiMo uses conservative 200-character chunks and rejects incomplete generation instead of silently advancing. This is a client-side limit, not an advertised provider maximum. Pricing/promotions may change; no permanent free-service guarantee is made. See the official synthesis guide.
OpenRouter: choose a secret, model and voice, then enable processing. Azure: also set the Speech resource region and cloud. Check your resource/provider privacy and retention settings; the plugin does not certify a service as no-training or ZDR merely because a key is configured.
BYOK: one active configuration supports OpenAI-compatible speech, ElevenLabs, MiniMax or experimental OpenAI Realtime. Fill in the endpoint, model, voice ID and secret reference, then explicitly confirm transmission, billing, training and retention risks. Changing the endpoint/provider clears the secret reference and consent without deleting the stored secret. Model, voice and secret-reference changes require renewed consent. Chat-only endpoints are not speech endpoints.
The OpenAI-compatible example uses the coral voice. OpenAI has announced that gpt-4o-mini-tts will shut down on January 6, 2027; settings warn affected official profiles. Select OpenAI Realtime (experimental) separately, re-select a secret and authorize it; do not just replace the old Speech model ID. Existing settings are not migrated. Each segment is buffered and its generated transcript checked before playback or WAV export; wording differences in numbers can cause rejection. Live OpenAI account, CORS and mobile-device validation are still pending; no latency or verbatim audio guarantee is made. Setup and limitations · OpenAI's deprecation notice. Other compatible services control their own model and voice availability.
Custom BYOK and remote CosyVoice requests use HTTPS, reject redirects, omit cookies and have no unsafe fallback. The service must allow requests from the mobile WebView (CORS); some endpoints may be unavailable. Test your actual device and endpoint. An authorization checkbox is not a provider privacy guarantee, and revocation cannot recall already-sent text.
Voice/API references: OpenAI speech, ElevenLabs, MiniMax.
Privacy
The default engine is device system speech. The plugin itself makes no TTS network request in that mode; the OS and voice implementation determine whether their speech processing is offline. Prefer voices identified as local when privacy matters. Installed voices may not be exposed to the mobile WebView.
All online engines are opt-in. After consent, the plugin sends text within the chosen reading range for synthesis. Enabling lookahead or continuity may also send upcoming text or neighboring sentences within that range before you hear them. API keys are read from Obsidian SecretStorage; only secret names are saved in data.json. OpenRouter requests require provider.zdr: true and provider.data_collection: "deny"; unavailable routes fail rather than silently weakening those flags.
The plugin has no developer relay server or built-in telemetry. By default, playback audio is kept in session memory. If you enable Reuse audio on this device, eligible audio is also stored on the device outside the vault, with size limits and expiry cleanup. This separate cache is not encrypted by the plugin; audio can contain private content, and OS backups may include it. You can clear it in Privacy settings. See Playback and Limits for retention and cleanup details.
Settings, optional resume records and explicitly exported audio stay in the vault and may be synchronized by your chosen sync service. OS, Obsidian and provider policies are separate. Stopping/timing out cannot undo processing or billing already begun. Failed requests are not automatically retried, and raw provider error bodies are not displayed.
OpenRouter defaults to Kokoro bf_emma (UK English female); presets also include Chinese and US English voices. Selecting another model chooses its own default, while saved voice choices remain unchanged on upgrade. Microsoft compatibility voice IDs may be rejected by a route even when accepted elsewhere. Models, voices, account access and prices remain provider-dependent.
PDF parsing and reading-order calculation happen locally. Image-only PDFs require OCR before they can be read.
PDF Sections and Footnotes
On Markdown, local HTML or PDF files, tap the outline icon or choose Document outline in the expanded player. Tap a title to locate the source and close the outline without changing playback. The adjacent range icon selects a section for Read section or Read from section. Section reading includes its subsections and stops before the next same-level or higher-level heading. HTML uses h1-h6 headings and needs HTML Reader for navigation. PDF bookmarks take priority over inference; unsupported exact navigation falls back to the page with a notice. Unresolved HTML navigation keeps the panel open rather than guessing a location.
Reopening the same unchanged PDF uses a single-document memory cache. Refresh explicitly rescans it; file changes or header-filter changes invalidate it. The cache is not written to settings and is cleared on plugin unload. Nothing is written back to the PDF.
Academic > PDF footnote reading defaults to Body only. Smaller bottom-page text with a marker and a separating gap may be identified as a note; uncertain text stays in the body. Choose Original order to retain notes, or Footnotes only to read them separately. The toolbar scope menu also offers Read PDF footnotes only without changing the setting. Selection-only reading preserves the selected text. Notes are separated by page layout, not linked semantically to individual citations.
Local HTML
Install and enable the HTML Reader community plugin to open .html or .htm files in the vault. Full-file reading parses the saved HTML locally, without executing scripts or loading page resources. For an accessible HTML Reader frame, selected-text and selected-position reading use the actual rendered selection. A reloaded frame or changed file may require selecting again. Script-generated content that is absent from the saved file is not included in full-file reading; external webpages and HTML highlighting are not included.
Audio Export
Start a reading range with an online engine, expand the player and choose Export reading audio (also available as a command). Confirm the displayed chunk/character count. Export synthesizes that entire range again, not just the remaining audio, so it may incur additional provider charges. Playback is paused; existing service consent still applies.
The result is a normal-speed, mono, 16-bit WAV under Note Reader Audio/ in the vault. Player speed and volume are not baked into it. No desktop executable is required, no files are overwritten, and exported content may be synchronized by your vault sync service. Close the export window to cancel future requests; already-sent requests and an already-started file write cannot be recalled. System speech cannot be exported. Mobile limits are 30,000 characters and 32 MiB output; choose a shorter selection or section if needed. Oversized/failed synthesis produces no completed export, but requests already sent may still be billed.
Remote CosyVoice contract
The configured endpoint must use HTTPS and accept a JSON request:
{
"input": "Text to synthesize",
"voice": "optional voice name",
"speed": 1,
"response_format": "mp3"
}
It must return MP3 or WAV bytes. If a remote secret is selected, it is sent as an Authorization: Bearer header.
Playback and Limits
Session audio and continuity: already prepared audio is reused within the same reading session (up to 64 parts / 16 MiB of cached audio bytes, excluding active responses and playback). In Playback settings, Prepare the next audio part is off by default. Enabling it permits one upcoming part to be synthesized alongside the current one, with at most two synthesis operations in flight. Future text may be sent and billed even if you never listen to it. Rapid jumps prioritize the latest queued target; sent requests cannot necessarily be cancelled and may still delay it. Pause prevents new queued work. Stop, changing documents or changing the voice/provider clears reusable session audio; exports and saved positions are untouched. Privacy settings also provide Clear session audio.
Optional device audio reuse: Privacy > Reuse audio on this device is off by default. Online/self-hosted audio can be reused across sessions for up to 24 hours / 64 MiB / 128 parts, with an 8 MiB limit per part. IndexedDB storage stays outside the vault; records contain audio, numeric metadata and hashed keys, not note text or credentials. Audio can still contain private content and is not encrypted by the plugin; OS backups may include it. Expired entries are removed on use, enabled startup and every ten minutes while enabled. Disable the option or use Clear reusable device audio to remove this separate cache. Storage failure falls back to normal synthesis and may incur charges. System speech and exports do not use it.
Experimental background audio: Playback > Advanced options > Background audio only is off by default. For online/self-hosted audio it avoids deliberately pausing when the app is hidden, stops hidden UI redraws, and exposes supported system play/pause, stop, seek and segment controls. Lock-screen metadata uses only the plugin name. It does not keep iOS alive or guarantee lock-screen continuity; system speech still pauses. Physical iPhone/iPad testing is required before relying on it for long background sessions.
Waiting-time diagnostics: run Copy playback waiting-time summary to copy bounded numeric statistics from this plugin session. Nothing is uploaded or written to disk automatically; the report contains no text, paths, voices or keys. It measures preparation, queueing, foreground waiting and the browser's playback-start event, not physical speaker output. sessionToPlaying starts when the reading queue is ready, not at the original click; extraction, when measured, is separate. Paused time is excluded from jump/gap intent measurements. First-time synthesis and slow providers still require waiting. Developers can run npm run benchmark:playback for a network-free synthetic comparison, not a real-provider speed claim.
In concise mode, simple formulas such as $z_{\mathrm d}$ are read as "z d"; verbose mode retains "subscript". Font and spacing commands do not make a short formula complex; explicit skip settings and complex-formula safeguards still apply.
Standalone \max and \min are read as "max" and "min"; common operators such as \lim and \sum and simple calls such as \sin x and \ln(x) are also supported. Unsupported or nested expressions and the explicit skip-all setting retain their existing behavior. Changes apply to a new reading session.
Concise style reads (x-y)/\bar z as "the difference of x and y, divided by z bar" and x/(y+z) as "x divided by the sum of y and z", including equivalent \frac forms. Verbose style, complex-formula safeguards and original highlight coordinates remain unchanged. Local processing measurements; phone hardware verification is still pending.
Conventional bounds such as [\ell,u] and matching E_{\min} / E_{\max} endpoints use concise "to" phrasing; verbose mode retains "closed interval". Simply subscripted adjacent factors are separated by "times".
Concise style recognizes short intervals when nearby prose explicitly identifies them or the formula uses membership such as x\in[a,b]. Open/closed endpoints stay distinct; unary signs use "minus"/"plus". In concise style, otherwise ambiguous short (a,b) or [a,b] pairs use neutral pauses ("a, b"), not an assumed interval; verbose style keeps brackets. Existing standalone numeric closed intervals and conventional bounds remain supported, but explicit coordinate/vector context prevents range wording. Simple two-argument calls state their arguments. Whole expressions such as (a+b)c use "the sum of a and b, times c"; isolated or unsupported arithmetic groups keep structural cues rather than silently losing precedence. Short literal sets use set/member labels (up to three simple members); PDF literal braces require math-font evidence. Complex smart-mode omissions and explicit skip settings remain in force. Original text and highlight coordinates are unchanged; start a new reading session for the new pronunciation.
PDF headers and footers: Settings > Academic > Skip PDF headers and footers is on by default. It locally removes short repeated edge lines (such as authors and journal names) and recognizable page numbers, without modifying the PDF. It applies to full-PDF reading, continuing from a selection and resuming; selection-only reading keeps the selected text. Changes take effect on the next reading session.
Filtering is conservative: uncertain edge text, larger titles and numbered headings are retained. When starting partway through a PDF, up to three preceding pages provide local recognition evidence but are not read aloud. Unique headers, unusual margins, rotated pages or missing text coordinates may remain unfiltered; disable the setting if body text is omitted.
- Online synthesis uses normal speed; playback rate is applied locally, avoiding doubled speed or a new paid synthesis merely to change speed. System speech rate/volume changes may apply at the next utterance.
- The five-second buttons and current-audio slider only seek within the loaded online audio, not future chunks or system speech.
- PDF extraction finishes before playback; this is not the desktop progressive-PDF implementation. Image-only PDFs need OCR.
- Highlighting, PDF bookmark editing and Web viewer/Copilot integration are not included in this release.
- Reading-order and resume anchors are heuristic. Complex PDFs and changed files may need a new selected starting point.
- Background/lock-screen playback is not guaranteed. Hidden playback pauses by default; the experimental opt-in applies only to online/self-hosted audio.
- If the device blocks autoplay, tap Resume to play the retained audio; this does not request another synthesis.
Mobile and Desktop Editions
This is a separate plugin and does not modify or replace the desktop edition. The table describes supported features, not completed device testing; see Compatibility and Testing for validation limits.
| Capability | Mobile | Desktop |
|---|---|---|
| Markdown and text PDFs | Supported | Supported |
| MiMo and custom speech APIs | Supported; endpoint compatibility varies | Supported |
| Local executable / Python engines | Not included; use an HTTPS self-hosted service instead | Supported |
| Academic text cleanup | Configurable Markdown and formula handling | Supported |
| PDF header/footer filtering | Default-on local heuristic; can be disabled | Supported |
| PDF outlines and footnote separation | Supported | Supported |
| Local HTML and audio export | HTML Reader integration; WAV export from online engines | Supported |
| Highlights, PDF bookmark writing, Web viewer, Copilot | Not included | Supported |
Manual Installation
Use this alternative when the community directory is unavailable or you need a specific version.
- Download
main.js,manifest.jsonandstyles.cssfrom the same version in GitHub Releases. - Put the three files in
<vault>/.obsidian/plugins/note-reader-mobile/using your device file manager or vault sync. - Enable the plugin in Community plugins. If updating an enabled plugin manually, disable and re-enable it to load the new files.
Do not copy another vault's data.json or credentials. Community approval does not replace testing on your device or with your documents.
Development
npm ci
npm test
This repository consumes @laginae/note-reader-core. It contains no Node.js, Electron, filesystem, child-process, or local executable calls.
See the device checklist for confirmed and pending coverage. Mocked tests do not send notes or charge speech accounts. The source entry point is included so the release bundle can be rebuilt from this repository.
Feedback
Report problems in GitHub Issues, including plugin/Obsidian/iOS or Android versions, speech engine, steps to reproduce and the relevant reading scope. For PDF issues, screenshots of the reader and the corresponding PDF region are particularly helpful. Redact personal content; attach a minimal PDF only if you have permission to share it. Never include API keys, secrets or data.json. Basic iPhone success is not a guarantee for every device or PDF.
Related repositories
- Shared core: https://github.com/laginae/note-reader-core
- Desktop edition: https://github.com/laginae/note-reader-cosyvoice
License
MIT