MushyBook
diction logo

diction

iOS keyboard that transcribes speech to text in any app

GoDeveloper ToolsGoMIT 56 7Health 68
GitHub

What is it?

What it is

An iOS keyboard for voice and AI that transcribes speech to text in any app, featuring an open-source Go-based gateway for self-hosting.

Why it exists

To provide a private, unlimited, and model-agnostic way to dictate, compose, and edit text via voice without switching apps or relying on proprietary cloud services.

Who should use it

iOS users needing voice-to-text in apps, privacy-conscious individuals, developers wanting self-hosted solutions

Who should avoid it

Non-iOS users, those not requiring voice transcription, users uncomfortable with self-hosting

How it works

A quick walkthrough in plain English

How diction works

Step 1 of 3

You interact with it

Open diction, send a request, or connect it to your stack.

Features

iOS voice‑to‑text keyboard that streams audio via WebSocket
Self‑hosted gateway written in Go handling encryption and model routing
AES‑256‑GCM end‑to‑end encryption with X25519 key exchange
Model‑agnostic: supports any Whisper‑compatible backend or custom speech server
Optional LLM cleanup for cleaner transcripts
Zero tracking – no analytics or telemetry
Docker‑compose ready for one‑command deployment
OpenAI API compatible endpoints for easy integration
Supports multiple Whisper models (small, medium, large‑v3‑turbo) and NVIDIA Parakeet
Cloudflare Tunnel, Tailscale, ngrok support for remote access
NixOS flake for non‑Docker deployments
Built with security hardening (DynamicUser, syscall filtering)

Advantages

  • Open‑source and MIT licensed
  • High privacy – no data collection or third‑party SDKs
  • Flexible deployment: Docker, NixOS, or custom backend
  • Free and unlimited usage for self‑hosted/on‑device modes
  • Encrypted data in transit
  • Compatible with existing OpenAI‑compatible clients
  • Easy to swap models via Docker Compose
  • Optional LLM cleanup improves transcript quality
  • No rate limits or caps on usage
  • Supports GPU acceleration for faster inference

Disadvantages

  • Requires iOS 17+ and full‑access keyboard permission
  • Self‑hosting needs Docker or NixOS setup
  • Large models need significant RAM or GPU
  • No TTS or word‑level timestamps support
  • No SSE streaming on REST, only WebSocket
  • Limited to Whisper‑compatible backends (no custom speech APIs without adaptation)
  • Requires network connectivity for self‑hosted mode
  • Potential latency on CPU‑only setups
  • Configuration can be complex for beginners

Installation

cloud

Install [Diction](https://apps.apple.com/app/id6759807364) on your iPhone.

docker

1. Create a folder for the stack and save the provided `docker-compose.yml`.
2. Run `docker compose up -d`.

native

nix run github:omachala/diction#diction-gateway

compose

1. Create a folder for the stack and save the provided `docker-compose.yml`.
2. Run `docker compose up -d`.

FAQ

What is diction?

diction is an open-source project licensed under MIT.

What language is diction built with?

Primary language: Go.

Is it free to use?

Yes. Licensed under MIT. Check the license for commercial use.

Loading documentation…
View on GitHub

Featured in Videos

YouTube tutorials and walkthroughs for diction

Alternatives

Similar projects ranked by category, topics, and text overlap.

Compare
ollama logo

ollama

Go

Run open LLMs locally with one command. CLI and REST API included.

177.1k·Health 90

Compare
HomeBox logo

HomeBox

Go

HomeBox offers a blazingly fast, Go‑based web application that runs in a single container. With an embedded SQLite database and responsive UI, users can organize items by categories, locations, and tags, upload photos, track warranties, and monitor purchase and maintenance dates. The project emphasizes simplicity, portability, and minimal resource usage (under 50 MB idle memory). It can be deployed via Docker Compose or run directly from container registries, with rootless and hardened images available. Community contributions, translations, and demos are actively supported.

6.6k·Health 85

Compare
ignis logo

ignis

JavaScript

Ignis is an actively developed open-source project that runs Obsidian in a browser via a server backend. It offers features like live sync, multi-vault support, and plugin compatibility while maintaining vault security through HTTPS and self-hosting. It is designed as a daily driver for note-taking with ongoing improvements.

874·Health 77

Compare
feishin logo

feishin

TypeScript

Feishin delivers a full‑featured music listening experience for self‑hosted servers. Users can install the desktop client (recommended) or run the web app via Vercel or Docker. The app supports MPV and web player back‑ends, automatic lyric fetching, scrobble integration, and a smart playlist editor for Navidrome. Configuration is managed through a GUI or environment variables, allowing server locking, authentication modes, and analytics control. The project is built with Electron‑Vite, uses TypeScript, and is distributed under GPL‑3.0.

9.1k·Health 85

Compare
muscriptor logo

muscriptor

Python

MuScriptor offers three transformer decoder-only variants (small, medium, large) trained on 170k songs, accessible via HuggingFace with a CC BY-NC 4.0 license. It provides a Python API, CLI, and a FastAPI-powered web UI for real-time transcription, piano-roll visualization, and MIDI export, with support for GPU/CPU, instrument filtering, batching, and beam search.

824·Health 77

Compare
diction | MushyBook