Product updates

News.Now shipped.

The latest development cycle closes the loop from installation to everyday interaction. Fine-grained fixes remain in the changelog; the five user-visible capabilities below are implemented and documented.

RELEASED

V5 Beta 5 · latency

Final recognition now prepares while you speak.

The old path could restart offline ASR and compile the current hotword graph only after voice-key release, leaving the entire cold start after even a long dictation. Beta 5 begins model, punctuation, and hotword preparation at physical key press and keeps the worker leased throughout recording.

macOS / IBus / Fcitx 5Recording overlaps cold startupLong dictation keeps the model aliveShort cold recordings await preparation

IBus and Fcitx 5 also begin microphone capture without waiting for a cold Core or user service; both paths in Universal packages receive the same fix.

Open the Beta 5 release →
RELEASED

V5 · macOS

VoCoType-linux is now native on macOS while keeping its historical name.

The original VocoType application crashed on the maintainer’s Mac, while the VoCoType-linux interaction model better matches the intended desktop workflow. Apple’s built-in dictation also did not fit this Chinese writing workflow: recognition was not reliable enough, continuing the same dictation after manual typing was awkward, and polishing, contextual voice editing, hotword replacement, and terminology protection were absent. V5 therefore ports this repository itself to InputMethodKit and AppKit instead of creating a separate product name.

Apple Silicon DMGInputMethodKit + AppKitDrag to ApplicationsConfiguration-preserving upgrades

Beta 4 also fixes first-run CoreAudio races, transient system clients stealing the text target, hotkey reloads swallowing F9 key-up, and status panes that could not be cancelled or dismissed.

Read the macOS guide →
SHIPPED

Live recognition

Native streaming ASR now updates preedit while you speak.

VocoType can optionally run the official FunASR online Paraformer in a local native worker. Partial hypotheses repeatedly replace the IBus or Fcitx 5 preedit during recording, including revisions to earlier text as later audio arrives.

Mutable preedit on IBus and Fcitx 5Local CPU-only native workerAutomatic fallback when preview failsFull offline model remains authoritative

The preview is deliberately non-authoritative: it is never committed. Releasing the recording key immediately sends the complete PCM buffer through the original Contextual Paraformer, hotword, punctuation, ITN, and normalization path.

Read the streaming ASR guide →
SHIPPED

Voice editing

Ctrl+F9 now speaks the same editing language on IBus and Fcitx 5.

Both integrations use one shared surrounding-text editing core. In compatible applications, spoken commands can replace, insert, delete, navigate, rewrite, format, undo, and redo without requiring Fcitx users to switch away from their existing Rime, Pinyin, Mozc, or keyboard input method.

One command model across both frameworksDeterministic and model-assisted editsSelection, cursor, and history awarenessClear capability checks and safe refusal
Read compatibility and limitations →
SHIPPED

Desktop setup

One graphical settings center replaces manual setup.

VocoType gives IBus and Fcitx 5 users the same graphical installation and repair workflow. General settings, recognition, terminology, AI, tutorials, and Playground tools are organized in one application; desktop Polkit dialogs handle the few operations that require system privileges.

Install and repair both integrationsDownload the runtime and ASR modelsManage recognition, terminology, ITN, and AITest audio, ASR, polishing, and editing
Read the settings center guide →
SHIPPED

Supportability

Doctor turns “it does not work” into actionable evidence.

The diagnostics flow checks the native runtime, dependencies, models, installed-file integrity, local and available versions, Fcitx modules, IBus components, services, IPC sockets, configuration, microphones, and the active processing pipeline. Users can then submit a redacted support bundle through the official feedback service or open a prefilled GitHub issue as a fallback.

Automatic environment and integrity checksVersion status and repair guidanceMicrophone waveform, recording, and playbackPrivacy-safe support bundle and feedback
Open troubleshooting docs →

Try the complete workflow, from installation to live editing.