VoCoType native C++ core

vocotype-core is the default speech backend for packaged Fcitx 5 and IBus installations. It preserves the existing Unix-socket JSON protocol, so the Fcitx module and the native IBus engine share one implementation of ASR, normalization, polishing, and voice editing.

Implemented

• Final Contextual Paraformer ASR through the official FunASR C++ runtime.
• Dynamic contextual hotwords, optional VAD, punctuation, and automatic audio

resampling in a disposable offline worker process.

• Native Chinese ITN with the existing context guards and 950 protected fixed

phrases.

• Live terminology YAML reload, longest-match canonicalization, protected

spans, legacy dictionary compatibility, and native hotword generation.

• Two-pass online preview through the official native streaming worker.
• F9 synchronous transcription.
• Shift+F9 asynchronous start/poll/cancel with OpenAI-compatible SSE deltas,

thinking suppression, heartbeat-aware idle timeout, and JSON fallback.

• Synchronous and asynchronous voice editing with strict JSON-plan validation,

key and modifier allowlists, repeat clamps, and a 32-action safety limit.

• Bounded task retention, recording cleanup, worker idle exit, signal handling,

typed configuration, fake-worker tests, socket tests, and real-model smoke tests.

Framework integration

• Fcitx 5 uses a global native module and launches the persistent user Core service on demand.
• IBus uses a compiled IBusEngine with direct librime integration.
• Both integrations use the same native socket protocol, recorder, ASR workers, terminology, ITN, SLM client, and voice-edit planner.
• The installed runtime has no Python backend or fallback path.

Build and test

make cpp-core-test

The FunASR worker bundle is built separately:

./src/workers/funasr/build.sh

Run an isolated development socket:

build/native-core/vocotype-core \
  --enable-final-asr \
  --socket-path /tmp/vocotype-cpp-test.sock