Speech To Text

v0.1.1模型接入

opencode-speech-to-text

Speech-to-text plugin for opencode — push-to-talk dictation via sox + Groq Whisper (Ctrl+')

维护活跃

#groq #opencode #opencode-plugin #speech-to-text #voice #whisper

GitHub 仓库 npm 页面官方主页

GitHub 星标

月装机量

385

近 7 天 175

综合评分SCORE

32.0

生态多维模型

快速安装与配置

opencode.json

写入当前项目的 opencode.json，只对这个仓库生效。

opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-speech-to-text@0.1.1"]
}

写入 ~/.config/opencode/opencode.json，对所有项目生效。

~/.config/opencode/opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-speech-to-text@0.1.1"]
}

opencode 启动时会通过内嵌运行时自动加载 npm 依赖并缓存至本地目录，无需手动在全局环境执行安装。

README.md

查看原始文件→

Push-to-talk speech-to-text for opencode — record with Ctrl+', transcribe with Groq Whisper, and drop the cleaned text straight into your prompt.

Features

Push-to-talk recording via sox (Ctrl+' toggles start/stop)
Groq Whisper transcription (OpenAI and custom endpoints too)
LLM cleanup pass — fixes punctuation and coding homophones (pie→pi, Jason→JSON, cash→cache, …)
Languages — auto-detect or pick one of several
Mic picker + persistent preferences across sessions

Prerequisites

sox — brew install sox
GROQ_API_KEY — get one at console.groq.com

Install

The simplest way — let opencode wire up both targets and install it:

opencode plugin opencode-speech-to-text -gf

This is a TUI plugin, so its real entry must be registered in tui.json (not opencode.json). To wire it manually, add it to the plugin array in tui.json:

// ~/.config/opencode/tui.json
{
    "$schema": "https://opencode.ai/tui.json",
    "plugin": ["opencode-speech-to-text"],
}

Restart opencode after installing — TUI plugins load at startup.

Usage

Press Ctrl+' to start recording, speak, then press Ctrl+' again to stop. The transcription is cleaned up and appended to your prompt. You can also run /voice (works in any terminal, even when Ctrl+' can't be transmitted).

Change the shortcut

Ctrl+' needs a terminal that transmits modified keys (kitty/CSI-u protocol). If yours doesn't, override the key via plugin options:

// tui.json
{ "plugin": [["opencode-speech-to-text", { "keybind": "ctrl+g" }]] }

Commands

Command	Description
`/voice`	Toggle recording (start/stop)
`/voice-cancel`	Cancel the current recording
`/voice-provider`	Switch transcription provider
`/voice-model`	Pick the Whisper model
`/voice-mic`	Pick the audio input device
`/voice-language`	Set language or auto-detect

Providers

Provider	API key env	Default model
Groq	`GROQ_API_KEY`	`whisper-large-v3`
OpenAI	`OPENAI_API_KEY`	`whisper-1`
Custom	(configurable)	`whisper-large-v3`

Cleanup runs on Groq Llama 3.3 70B (falls back to 3.1 8B) and also uses GROQ_API_KEY.

Development

pnpm install
pnpm run build
pnpm run lint

License

MIT