233stone/vocotype-cli

VocoType CLI – Offline Speech‑to‑Text Engine

What it is

• An open‑source command‑line version of the VocoType speech‑input system. It runs entirely on the user’s computer, converting spoken words to text in real time without sending any audio to the cloud.
• Built on top of the FunASR offline recognizer (Alibaba/Damo) and optionally able to use Volcengine’s BigASR cloud service.

Core capabilities

Feature	Details
Offline inference	Uses a ~500 MB FunASR model that runs on CPU only (≈700 MiB RAM). No GPU required.
Fast response	Claims ~0.1 s latency for local inference.
Chinese‑English mixed transcription	Handles bilingual input, useful for technical terms.
AI‑driven post‑processing	After raw ASR output, a lightweight AI model (selected via prompt templates) corrects homophones, typos and spoken edit commands like “改成 …”.
Custom replacement dictionary	Up to 20 user‑defined terms (e.g., names, jargon) that are forced into the final transcript.
Optional cloud backend	Can be switched to Volcengine BigASR by providing an App Key and Access Key in config.json.
Dataset export	--save-dataset flag writes the captured audio‑text pairs to disk for further training or analysis.

Typical users

• Professionals who need a privacy‑preserving dictation tool (writers, lawyers, doctors, researchers).
• Developers who want a programmable speech‑input component for their own apps.
• Anyone who prefers a CLI that can be scripted or integrated into custom workflows.

Installation & quick start

# Clone the repo
git clone https://github.com/233stone/vocotype-cli.git
cd vocotype-cli

# Create a Python 3.12 virtual environment (uv is recommended)
pip install uv
uv venv --python 3.12
source .venv/bin/activate   # Windows: .\.venv\Scripts\activate

# Install Python dependencies
uv pip install -r requirements.txt

# First run – the script will download the ~500 MB FunASR model automatically
python main.py

• To record and transcribe, simply run python main.py and speak into the default microphone.
• To export the captured audio‑text pairs: python main.py --save-dataset.
• To use the Volcengine cloud recognizer, create config.json with the credentials shown in the README and launch with python main.py --config config.json.

Where to get the full desktop experience The CLI is aimed at developers. Non‑technical users are directed to the free graphical VocoType desktop client (download links for Windows/macOS are provided in the repo’s README).

Project health

• Releases are tagged (e.g., v1.6.0) and include pre‑built installers for Windows and macOS.
• The repository lists its main dependencies (FunASR, optional Volcengine SDK) and provides clear setup instructions.
• Issues and feature requests are handled via GitHub Issues.

Why it matters VocoType‑CLI demonstrates that high‑quality speech recognition can be delivered locally on modest hardware, offering a privacy‑first alternative to cloud‑only dictation services while still allowing an optional cloud backend for users who need the extra accuracy of large‑scale models.