8.9 Browse Installed Voices And Choose Engines

Open C-e H v for Voices and engines. b browses engines and voices, p palettes, and w resumes Voice Workbench. Browser q commands unwind child views and return to the originating menu item. Filters, selection, and staged work survive; unsaved engine-policy changes are still announced. A directly opened browser retains its ordinary return destination.

The main Workbench views are:

e

Engines, for trying voices or inspecting availability.

v

Physical voices, including engine voices and Omnivox downloads; on an engine row, restrict to that engine.

l

Named voices, labelled Logical voices, such as bolden and annotate.

s

Styles and effects for those entries, including supported parameters.

From the engine browser at C-e H v b, v opens voices. RET on an Omnivox engine opens status, downloads, and settings; other adapters use it to browse voices directly. RET or P on a voice auditions, x shows details, and t opens an experiment. q returns through the engine’s details and list as appropriate before leaving Workbench.

Filters set with F and each view’s row/column survive navigation through settings menus. The browsed engine is a temporary limit, removed when switching to logical, style, or engine views, or pressing v from a voice list. Explicit filters remain; C clears them to show voices across engines. Direct view shortcuts remain available.

8.9.1 Find and filter installed voices

A language with one voice shows it directly; several voices form an expandable heading with count and state. RET toggles it. Initial sorting ignores capitalisation. Regions form a tree, for example English then United Kingdom then English (West Midlands). Redundant single-child levels are omitted; private language-tag codes do not create extra headings. Unknown codes stay visible and missing language is grouped as Unknown language.

A one-language list starts expanded. Each engine scope remembers folds; refresh preserves voice and column as library data arrives. Sorting keeps headings with their voices. A samples usable visible voices, including single-voice languages, in display order. Expand a group to include hidden voices.

/ searches names across collapsed languages and disabled downloads within the current engine and explicit filters. Choosing expands its language and selects it. A hiding quick filter switches to All voices. Search neither plays a sample nor changes enablement.

Omnivox quick filters are available through Q with counts, or directly:

1

All reported engine and installed library voices.

2

Catalogue downloads, including disabled ones; imports and built-ins remain in All voices.

3

Library voices enabled for the next Apply.

4

Needs Apply: desired enablement differs from either worker’s eligibility. Unknown or stale worker evidence cannot establish a pending change and is counted separately.

Counts include collapsed groups and respect engine/explicit filters. Each quick filter remembers row/column. After Apply empties Needs Apply, 2 returns to downloads for previewing or editing. Engine overrides may prevent confirmation; Active and engine details still report health and usability. C clears the quick filter as well.

8.9.2 Distinguish installation, enablement, and active speech

A voice can be installed without being ready to speak. Installed means it belongs to the library or is supplied by an engine. Enabled records whether it should be available after the next Apply. Active identifies eligibility on the main stream, notification stream or both; it does not say that a model is currently resident in memory.

Library information arrives asynchronously. When worker evidence is missing or inconsistent, the view reports Unknown rather than guessing whether the voice is active.

+ enables/disables a managed voice, a reviews Apply, and d opens downloads. Downloads start disabled; enable and Apply before auditioning. Engine-managed voices retain their own selection mechanism. Other adapters retain their physical-inventory view.

To remove a downloaded Piper, Flite, or MBROLA package, disable its voices, Apply, then press C-c u on one voice. Review all speakers sharing it, size, known palette references, and retained-file reasons. Confirmation removes only reviewed managed files. Built-in voices, imports, and runtimes remain, as do saved palette choices for a later redownload. The backend guide describes supported managed-engine download and removal boundaries in more detail.

Another session may still need the package; files remain until retirement is confirmed, and older session records may need inspection. Results distinguish removed bytes, remaining files, and unconfirmed amounts. C-c U or omnivox-library-resume-uninstall reviews interrupted cleanup on hosts supporting uninstallation.

Check an engine’s Availability before changing its routing policy. A missing runtime remains listed as unavailable, with Last failure explaining the problem. Routing policy: allowed permits selection; it does not install the runtime. An engine being disabled, unhealthy or temporarily failed are also distinct conditions.

8.9.3 Compare samples and interpret the result

A auditions available matching voices in displayed order, skipping unavailable rows. B uses the selected voice as A and asks for B; Search all engines can compare outside current filters. Both use one sentence. T edits it, F filters, C clears filters, and R refreshes inventory.

Use C-e H v o l to choose comparison labels: engine and voice names, letters or sequence numbers, or no labels. The option is emacsvox-aural-preview-label-verbosity. The list starts on the current choice. Arrows browse without applying; RET applies and returns, while q cancels. This global option changes labels, leaving sample text and order intact. With no labels, the sequence no longer identifies its voices aloud. Press w in Voice presentation to save labels and pitch contrast through Customize.

Labels and samples form one sequence: S or foreground navigation cancels remaining playback. Exact-preview failure is reported rather than accepting a substitute as the requested voice. Capabilities describe adapter claims, not a listening test of every parameter.