Organize recordings and create a report across records

Projects, folders, entries and takes

Open the project library in the session browser under Meetings & Minutes. Select a name to choose a location, or use its arrow to expand it. The older chronological list remains available in a collapsible section.

123

1. 1Expand an entry, then open a take to review its existing transcript, speaker and emotion records.
2. 2Submit recording start from the selected entry. Follow the main window for the actual state and Stop.
3. 3Review the selected take list. This step does not generate a report.

Takes reuse the existing detail and editing area. Use the report's Transcript and original audio section below to listen to a cited recording range.

In the r490 source preview, click Clean up text on a transcript segment to bring its selection or full text into an independent copy. Compare the result before copying it; takes and report quotations remain unchanged. See the text cleanup workspace steps.

Organize existing records first

1. Choose New project, enter a name and save. Select the project and use New folder to create a group.
2. Expand Unclassified originals. Find an older recording and choose Organize. Name the entry, choose its project, folder and type, then save. The original recording stays in place.
3. To add another saved recording to the same entry, select the entry and choose Add take. Select an unclassified original instead of importing or copying the audio again.

Use Rename on a selected project, folder or entry. Use Move to change a folder or entry's location. After archiving, turn on Show archived, find the item and choose Unarchive. Restore an archived parent project or folder first.

Find remembered words across recordings

Use Names when you remember a title. Switch the same search box to Transcripts when you remember spoken words. Transcript search reads local organized, unarchived transcripts. It does not call models, read audio or upload material.

1. Select a project, folder or entry, then switch to Transcripts. Check Transcript scope for all projects, the current project or a folder. A folder scope includes its subfolders.
2. Enter a continuous fragment of the original wording. Search starts automatically; Enter starts it immediately. Matching is literal, with ASCII letters matched regardless of case. It does not find synonyms or similar meanings. Try a shorter original phrase if nothing matches.
3. Check the number of takes reviewed and matching passages. Choose Stop search or press Esc while searching. Completed results show up to 20 passages per page. Use Previous and Next to browse.
4. Check the project, folder, take, speaker and time shown with a result, then choose Open original passage. The app verifies the current source before opening its transcript segment. Audio does not play automatically.
5. To use related material in a report, check Report selection (whole entry). After verification, the entire entry joins your existing selection, including all its takes. This does not select only the matching sentence. Continue with report scope, review sources, then freeze and generate.

Check source coverage before treating an empty result as final

Transcript search excludes unclassified and archived entries. Organize the record or unarchive it before searching. Changing modes, stopping or retrying a search retains records, report selections and editing drafts. Report generation still requires your separate confirmation.

Start the next recording from an entry

1. Prepare the microphone and recognition model in the Setup guide. You can create an empty entry by leaving its original-session selection empty.
2. Select the entry. Choose Start recording when it has no takes, or Continue recording when it already has takes. This request keeps its original target even if you select a different entry afterward.
3. A submitted-start notice means the request was accepted. Check the main window for the actual recording state. The library start button stays disabled during preparation, recording and background finalization.
4. Choose Stop in the main window. Wait for saving and assignment to finish, then expand the entry to check the new take. Earlier takes remain separate.

If start fails, use the main window's error message to check your configuration and retry. A failed submission does not display a running recording or automatically start another one.

Choose the records to summarize

• One entry with several takes. Select the entry and choose Generate Linglu report.
• Several entries. Check their boxes and choose Generate Linglu for selection.
• A whole project. Select the project or one of its entries and choose Generate Linglu for whole project.

You can first use Review entry sources, Review selected sources or Review project sources to inspect the take list. Those buttons check organization only. To freeze transcript contents and generate a report, use Linglu · Project reports. Changing the library selection does not automatically freeze, upload or generate anything.

Freeze the sources, then generate the report

1. In Linglu · Project reports, check the scope name and choose a report title and language.
2. Choose Source inclusion. The default, All sources (all must be ready), requires every selected transcript source to be ready. If transcription is unfinished or a source is missing, wait for it to finish. You can instead explicitly choose Ready sources only and choose Freeze and review sources again. The app does not silently switch to a partial selection.
3. Review each ready, unfinished or unavailable source under Review frozen sources, along with archived entries and entries with no recording takes. “All” means the selected transcript sources are complete; it does not promise that all audio is playable. A transcript can still be included when audio is missing, with playback marked unavailable. Partial scope is labelled and its omissions remain listed.
4. Check the selected meeting model. Use Manage meeting model to change the existing meeting configuration; there is no second address or API-key form here. Freezing saves a local snapshot and does not generate a report. After reviewing it, choose Generate report separately. Only then are included sources sent to the selected meeting model.
5. Follow the stage and progress, or choose Cancel report task. Changing entries or hiding the view does not change the running task's scope; return to see the same background task. A failure or cancellation keeps the sources and your draft. Retrying requires another explicit click.

Under Saved reports, select a title to open a local report without generating it again. Choose Export report HTML and select a destination. The export preserves report text and source references; recording files are not attached automatically.

Voices Remembered · turn recorded conversations into a memoir

These memoir steps require the Windows r512 or later installer preview. Select one entry, several entries or a whole project and freeze its sources using the steps above. Recordings stay in place and the existing meeting model is reused.

1. Check the frozen selection and omitted material in Project reports. Under Report format, choose Voices Remembered · memoir. Under Chapter arrangement, choose Chapters in selected source order or Chapters by theme.
2. Source order follows the selected recordings. A thematic arrangement groups related subjects. A recording date says when a conversation was captured; it does not establish when the remembered event happened. Keep unknown event dates uncertain.
3. Confirm the title, language and existing meeting model, then click Generate report. Wait for completion before reviewing each chapter and its citations. Click a chapter heading to collapse or expand it.
4. Click Edit memoir, then use Move chapter up and Move chapter down. Changes stay in the current draft until saved. They do not move recordings or rewrite quotations.
5. Change the Revision title, chapter headings or narrative text. Keep every existing chapter, paragraph and citation. Titles and chapter headings allow 160 characters each; a paragraph allows 1,600. Original quotations, source references and times remain read-only. Follow a citation to inspect frozen text and load original audio when needed.
6. After reviewing, click Save as new version. Edited narrative is marked Human revision and the previous report is retained. Use Open previous version or select it in Saved reports. Saving a version does not replace original audio or transcripts.

1234

1. 1Choose the memoir report format and chapter arrangement.
2. 2Expand chapters, edit the narrative and adjust chapter order.
3. 3Save a new version while keeping the previous report available.
4. 4Keep original quotations read-only and check source time and affect evidence.

Each narrative paragraph keeps its original citations. If a quotation is wrong, review the saved transcript, then freeze and generate again. Do not rewrite quoted speech inside chapter editing to fit the story.

When switching reports or finishing editing, unsaved changes prompt you to choose what to do. Save a new version to keep them, or explicitly discard the draft. If saving fails, the current draft and previous report stay available. Review the notice before trying again.

Continue using quotation, affect-evidence and audio checks. When saved metadata is available, the source view shows the recording start time; older material may not contain it. Model affect tags and human choices remain separate. Missing evidence does not justify adding feelings, psychological explanations or story details. Check matching names, conflicting accounts and unknown dates against their separate sources.

Export the current version as HTML to share it. The file contains narrative text and source citations; it does not package all recordings.

Keep saved independent affect observations with a report

These historical-observation steps require the Windows r512 or later installer preview. Model paths remain in Setup; reports do not need another configuration.

1. Select a segment in the meeting record and expand Dedicated emotion analysis. Choose the registered Emotion2Vec classifier, check an interval of 0.25 to 10 seconds within the segment, and click Analyze and save. Wait for the result to be saved with the meeting.
2. Return to Project reports and click Freeze and review sources. Freezing includes only independent observations already saved at that time. It does not start the affect model or analyze missing samples. To include observations saved later, freeze again and generate a new report.
3. After generation, expand Appendix · independent affect observations, then expand an analysis. The appendix lists saved observations only for segments cited in the narrative. Click Review this segment, then expand Emotion2Vec · saved acoustic observations in source details to inspect every frozen observation for that source. Use the page controls when there are more than 12.
4. Review the candidate class, nine raw scores and model revision alongside the transcript and audio. Each analysis retains its own interval, audio range SHA-256 and model SHA-256. Saving a memoir as a new version retains these observations; narrative editing does not replace earlier analysis records.

1234

1. 1The report appendix contains observations only for cited segments.
2. 2Source details retain all 14 frozen observations in this example, including uncited segments.
3. 3Expand an analysis to inspect its nine raw scores and provenance. The scores are uncalibrated.
4. 4The transcript model tag and human annotation remain separate from independent acoustic observations.

Independent observations appear in local source details and report appendices. They are not sent to the report model and do not establish identity, inner feelings or emotions in a remembered event.

Check whether observations were captured, then check the audio status

The report preserves the check performed at freeze. Later playback uses an actual interval of the current normalized recording; it does not authenticate historical observations again. To analyze new audio or include later results, analyze and save in the meeting segment, then freeze and generate again.

Review the report and original words together

This reading layout requires the Windows r512 or later installer preview. The Mac r492 source ZIP does not include it.

1. Click a source and time below a report paragraph. On wide containers the report and frozen transcript appear side by side; on narrow containers use Report / Evidence. The evidence pane also shows the corresponding report paragraph.
2. The selected segment and its transcript page are revealed automatically. The report, source and transcript keep their own reading positions. Use Return to citation or Close transcript to return to that citation.
3. Click Load segment audio only when you want audio. The player appears above the transcript and never starts automatically. It reads the same bounded actual interval as before; it is not a full-recording player.
4. While editing a memoir, the evidence pane shows your current unsaved paragraph. Reviewing evidence does not save or discard the draft. Expand Source time and playback details for recording metadata and the limits of current-audio verification.

Review quotations, affect evidence and recording ranges

1. Select a source name and time below a report paragraph to open that take's frozen transcript. Segments named “s1” remain separate across sources.
2. Compare the words, speaker label and time. Recognition transcripts can contain errors and are not human-verified ground truth. Review paragraphs labelled AI summary or Model observation; they do not establish identity or psychological state.
3. Inspect affect evidence. Model tag preserves the model observation. Human annotation separately shows the saved label, unknown choice or choice to keep the original model tag. They can differ. Neither is an intensity measure, and a human choice does not prove the model correct. For saved independent Emotion2Vec observations, follow the separate observation and audio checks above.
4. Choose Load segment audio when you want to listen. For a segment longer than ten seconds, load its first or last ten seconds, or enter a start time. Each request loads an actual 0.25 to 10 second range of the 16 kHz mono recording. A shorter final range is not padded. Nothing plays automatically; check the displayed times and press Play.

Playback uses the current normalized recording, not the original device file, and does not authenticate historical audio contents. Missing audio does not hide the frozen text. If the transcript was later edited, a notice says that the frozen version is still in use. Name withdrawals or privacy changes block the old source until the records are reviewed again.

Source coverage and omissions distinguishes cited sources, processed but uncited sources and sources not sent to the model. Absence from the report text does not mean a source was unprocessed or proven unimportant.

When a notice appears

• The library changed. Your draft stays in place. Choose Refresh, review the updated location and contents, then save. The new revision does not silently overwrite your edits.
• A recording was saved and its location needs review. Refresh and inspect the target entry. If the recording remains unclassified, use Add take to assign it manually. If it already has the correct location, keep it there. Preserve the original record.
• A saved-recording notice arrives while you are editing. The draft is kept and the library does not automatically advance to another revision. Refresh and review before continuing edits.
• The new take is not visible. Wait for the main window to finish finalization, then refresh. A submitted start alone does not confirm that a recording has been saved.

For a report from one meeting, continue using Meetings & Minutes or LingLu topic minutes.

Periodic retirement: keep expired export copies in one place

Requires Windows r513 or later. Installer previews r512 and earlier do not include periodic retirement. The original manual preview and confirmed-move workflow remains available.

1. Open Storage manager at the lower right of the project library and find “Cache retirement mode”. It is Manual by default. You can keep it manual and use the existing preview steps first.
2. For automatic checks, change “Retirement mode” to Periodic retirement. The choice saves automatically. Wait for “Saved”; this window can then be closed, while the app must remain running.
3. Select Read retirement status to see the next planned check, the latest round and the archive directory. A busy app, recording, exit or another active task defers a round. Failures back off before retrying, instead of scanning continuously.
4. Switch back to Manual at any time to stop scheduling future automatic rounds. Changing the mode does not interrupt a round already in progress. Read its status before manually handling the archive directory.

The default interval is 60 minutes, after a 5-minute delay following app startup or enabling the mode. Only verified newer export-check copies older than 30 minutes, without an active task reference or file lease, are considered. This is not a whole-disk cleanup.

No files are deleted, and moving files does not free disk space. Eligible copies move to the displayed “可删除” archive directory. You may handle that directory yourself after reviewing the results. Models, original recordings, history, cited report sources, pending or failed material, incomplete .zip.partial files and unverified orphan work copies are excluded from periodic retirement.

123

1. 1Choose manual or periodic retirement; changes save automatically.
2. 2Read status without starting a new scan or move.
3. 3Review the next check, latest round and archive directory; a move does not mean space has been freed.

Read a transcript and follow original audio

In Session browser → Transcript preview, each passage is displayed as readable text. Time, speaker, Clean up text and Edit sit together; short sentences do not reserve a large editor. These changes are included from r492.

1. Click Edit to change the passage or its speaker name. Use Save text or the appropriate rename button to save. Closing the editor keeps the draft; Restore discards the local changes.
2. Click a passage time to listen from that passage. Pause, Stop and Continue passages control playback. Missing or unreadable recordings leave the transcript available.
3. Follow audio highlights each timestamped passage. Manual scrolling or filter/page navigation pauses following; editing does not get interrupted by auto-scroll. Turn Follow audio back on when ready.
4. Starting recording or transcription stops original-audio playback. Changes to the recording or transcript require a refresh before listening again.

Highlighting follows passage timestamps, not individual words. Continue passages moves to the next saved passage start, omitting unselected gaps. Live, unsaved recording does not provide this saved-audio playback. Audio stays local and is loaded only when requested.

Reuse corrected terms within a project

Select a project, including an empty one, then choose Project dictionary and history to import and review its own terms. See the batch import and history guide. Available from r508. Source r510 also carries the dictionary and existing history with project migration.

After assigning a recording to a project, open Clean up text from its saved transcript. The workspace checks actual ownership and defaults to project-only terms. Moves, archives and dictionary changes require review; new rules do not rewrite original transcripts or existing reports.

Read the project dictionary walkthrough

Take an entry, selection or whole project with you

This entry is available with the r505 source preview. It exports saved text and recordings without running recognition, generating a report or uploading material.

1. Select an entry and choose Export entry material, or select a project and choose Export whole project. To include several entries, check them and choose Export selected material.
2. Choose TXT for readable text, SRT for subtitles, and JSON for timestamps, take associations, speakers and saved emotion records. These are not new emotion analyses.
3. Turn on Include saved audio copies when needed. Use Choose location to select a disk and folder. The parent directory of a successful export is remembered; each new export still needs review and confirmation.
4. Choose Prepare export. When the preview is ready, review ready takes, project dictionaries, saved reports, omitted items and reports, missing audio, estimated size and free disk space. Expand Review files and omissions; the file list scrolls independently.
5. Acknowledge omitted items or reports, then choose Start export. Use Change options to change formats or location and prepare again. Each export creates a new ZIP or folder; identical titles have separate numbers and existing exports are not overwritten.
6. Review the result and use Copy location to find it. Keep the complete ZIP, or the entire directory for folder exports. Audio paths are relative to its manifest.json, rather than absolute paths on the original computer.

If the original WAV is missing, invalid or too short, the session's normalized recording is checked and labelled separately. If neither is usable, text and the missing-audio notice remain. JSON keeps recognition-model emotion, human choices and saved independent acoustic observations separate; raw scores remain uncalibrated.

Closing the card keeps background work running. Open Export again to check progress. Exiting the app warns about unfinished work; background work needs the computer to remain on. A preview expires after 30 minutes in this app run. The r507 source preview can reimport this complete folder using Import project material below.

Bring an exported project back into the library

r507 source preview.Continue organizing an exported project on another computer or in another library. Only manifest-listed files in the chosen folder are read. No model calls or uploads occur; the export and existing local records remain.

1. Locate the complete Boundless Flow export folder with manifest.json, manifest.sha256, records and JSON. TXT or SRT alone cannot restore the project hierarchy. Export again with JSON if needed. To carry sound, the original export must include saved audio copies.
2. Choose Import material → Choose ZIP / Choose folder → Check import material above the project library. Checking reads the manifest, text and file checksums; it does not import yet. Finish any active recording or speech task first.
3. Review project names, takes, dictionaries, saved and omitted reports, carried audio, text-only takes, estimated space and free space on the library disk. Acknowledge omitted items, omitted reports or missing audio before continuing. Text-only takes retain their words, timestamps and saved labels.
4. Choose Import as new projects. Projects, folders, entries and takes receive fresh IDs, keeping take order without merging identical titles or overwriting recordings. Human emotion choices and saved independent Emotion2Vec observations remain separate. Historical observations retain their original date and model and are labelled “Imported saved”; no new inference runs.
5. The library refreshes on completion. An active editing draft remains; refresh the library before continuing edits. Open the imported takes and check text, timing and audio. Missing sound is not recreated from the text as “original audio”.
6. For the same complete export folder, Recover imported material returns the existing IDs without duplicating records or overwriting later edits. Finish previous import verifies copied files and finishes registration. This is an explicit retry, not automatic work after restarting.

Previews stay valid for 5 minutes in this app run. Closing keeps work running; quitting warns about background tasks and the computer must remain on. Checksums verify integrity, not authorship, speaker identity or authenticity. Source r511 supports this app’s ZIP packages directly, alongside complete versioned folder exports. New r510 JSON packages also carry project dictionaries and history, and available saved reports with frozen source text within the selection. Older folder packages still import transcripts and carried audio. See the ZIP and empty-project steps below. Arbitrary TXT/SRT files still cannot restore projects directly.

Carry dictionaries, reports and original words together

The Windows r511 or later installer preview includes these r510 project-migration features. Use the export and import steps above; no extra migration settings are needed.

1. Keep JSON selected when exporting. New packages automatically carry project dictionaries and existing history, plus available saved reports and the source text frozen when they were created. TXT / SRT alone do not carry these materials.
2. Review dictionary, report and omitted-report counts. Reports involving entries outside the selection, or unavailable reports, do not silently bring in material outside your scope. Expand the selection and prepare again when needed. Omitted reports require acknowledgement before continuing.
3. After import, open Project dictionary and history in the new project to review its terms and existing history. The original project and global dictionary remain unchanged. Open migrated reports under Saved reports without generating them again.
4. A report retains its original frozen text and citations, while the current transcript retains later saved edits. Review both when they differ; newer words do not replace historical quotations. Valid carried audio supports citation playback. Without carried audio, reports and quoted text remain readable.
5. An “older package” notice means dictionaries and reports were not included when that package was exported, rather than lost during this import. Export a new JSON package from the original library with the newer version to include them.

Keep the whole package together rather than copying only report or dictionary files. Failed verification is not shown as a completed report migration. Review the original library and export again.

Move one ZIP file, including an empty project

Available in the Windows r511 or later installer preview. Move to another computer, keep a project copy, or carry an organized hierarchy and terms before recording starts. No recognition, report generation or upload runs.

1. Select a project and choose Export whole project. Under Package format, keep the default ZIP file (recommended). Choose Folder when you want to browse individual files.
2. Keep JSON selected, choose a location and select Prepare export. Review dictionaries, reports, takes and omissions. An empty project has 0 takes and can still retain its structure and terms. Existing recordings without usable transcripts are explicitly omitted rather than treated as a complete empty-project backup.
3. Choose Start export. Keep the complete .zip file, or the whole directory for a folder export. Audio may not become much smaller: the ZIP keeps related material together so files are not missed during copying.
4. In another library, choose Import material → Choose ZIP → Check import material. No manual extraction is needed. Use Choose folder for older folder packages. Review projects, takes, dictionaries, reports, omissions and disk space, then choose Import as new projects.
5. After importing an empty project, add entries and recordings and keep using its dictionary. For the same package imported before, use Recover imported material to avoid duplicates while preserving subsequent edits.

ZIP import also needs temporary extraction space, included in the estimate. When the task completes or is cancelled, verified task-owned cache moves to zip-import-cache-v051/可删除 inside application data; it is not physically deleted or reused. Unverifiable cache is retained. Use Storage to find application data and handle the archive manually. Moving the archive still occupies its original disk; space has not been reclaimed.

Only this app's versioned project packages can restore projects; arbitrary ZIP, TXT or SRT files cannot. Checksums do not establish authorship or human identity. Review migrated reports and citations using the dictionary and report migration guide.

Inspect storage, then confirm cache retirement

Choose Storage in the main window, or Inspect storage usage beside the audio-cache location in Setup. Inspection uses saved settings, not an unsaved folder draft.

1. Choose Inspect storage. The app inspects file sizes in its own and registered locations. It does not load models, read audio contents or scan the system temporary folder. Model folders are optional: enable Include model folder sizes when needed.
2. Review recordings, report sources, recovery material, export cache, projects, pending or failed work, unverified chunks and models. Expand Locations to copy a folder. Missing paths, links and inspection limits are marked incomplete with lower-bound sizes rather than verified empty folders.
3. Check current free space for the app-data and audio-cache volumes. Use Change future audio cache location to save a new location in Setup. This affects new recordings and does not migrate existing material.
4. Choose Preview eligible cache. Only newer export-owned cache with complete ownership and content verification, older than 30 minutes and without an active lease is eligible. Failed, pending, changed or unverifiable material and old cited report sources remain.
5. Review the list, check the confirmation and choose Confirm archive move. A preview is valid for five minutes, and contents and leases are checked again before moving. For a bounded partial batch, review and process it before inspecting the rest.
6. Review the actual archived and preserved counts. The move does not reclaim disk space; the app does not physically delete the archive. Copy its location and decide when to handle it manually. Archived contents are not read or reused by the app.

Closing the window keeps background work running. Open it again to recover status. If status cannot be verified, use Read current task without repeating the operation. Use Cancel inspection to stop; an active operating-system disk read must return before cancellation finishes. There is no restart continuation or scheduled physical deletion.