Speakly

A lightweight floating TTS mini-player for your desktop.

Install

pipx install speakly

Or with Homebrew:

brew install noncuro/speakly/speakly

Quick Start

speakly "Hello, world!"           # uses free edge-tts, no API key needed
speakly                           # reads from clipboard
speakly --file article.txt        # reads from file
speakly --speed 2.0               # set playback speed

No API keys, no accounts, no configuration. It just works.

Features

• Floating mini-player -- always-on-top window with play/pause, scrub, speed control (1x-3x), and volume
• Free out of the box -- uses edge-tts (Microsoft Edge TTS), no API key required
• Premium voice upgrades -- swap in OpenAI, ElevenLabs, or Inworld when you want better quality
• Smart titles -- auto-generates a short title for the player window (Claude or GPT if configured, heuristic otherwise)
• Audio caching -- SHA256-based cache in ~/.speakly/cache/ with 7-day TTL; repeated text plays instantly
• Keyboard shortcut support -- wire up a system hotkey via macOS Shortcuts or Raycast
• Catppuccin Mocha dark theme -- easy on the eyes at any hour

Providers

Provider	Quality	Speed Control	API Key	Best For
edge (default)	Good	Player only	None needed	Getting started, casual use
OpenAI	Great	Native 0.25-4x	Required	Speed range, consistent quality
ElevenLabs	Excellent	Player only	Required	Multilingual, natural voices
Inworld	Excellent	Player only	Required	High-quality at 2-3x speed

Recommended: Inworld for fast listening

If you like listening at 2-3x speed, Inworld's TTS 1.5 Max model sounds excellent even sped up. Generation is fast and the audio quality is superb. Getting started takes a minute:

1. Sign up at inworld.ai
2. Grab your JWT key and secret from the dashboard
3. Run speakly config to save your keys

Configuration

speakly config    # interactive setup -- set provider, speed, API keys

Config is saved to ~/.speakly/config.toml. API keys are stored securely in your system keychain via the keyring library.

You can also set environment variables directly:

export OPENAI_API_KEY=sk-...
export ELEVEN_API_KEY=...
export INWORLD_JWT_KEY=...
export INWORLD_JWT_SECRET=...

Usage

speakly [OPTIONS] [TEXT]

Options:
  -p, --provider  TTS provider: edge, openai, elevenlabs, inworld   [default: edge]
  -v, --voice     Voice name or ID
  -s, --speed     Playback speed multiplier                         [default: 1.0]
  -f, --file      Read text from a file
  --list-voices   List available voices for the provider
  --help          Show help

Commands:
  config          Open interactive configuration

Keyboard Shortcut (macOS)

Three steps to read any text with a hotkey:

1. Download Speakly.shortcut from the latest release
2. Double-click to import into Shortcuts
3. Go to System Settings > Keyboard > Keyboard Shortcuts > Services and assign a hotkey (e.g., Cmd+Shift+S)

Then: select text anywhere, copy it (Cmd+C), press your shortcut -- Speakly reads it aloud.

Some apps that support macOS accessibility services (Safari, Notes, TextEdit, VS Code) let the shortcut grab selected text directly, skipping the copy step.

Performance

Synthesis latency measured on macOS (headless, no UI overhead). The player window itself appears instantly (<150ms) regardless of provider.

Provider	Text	1st Audio	Total
edge	short (~200 chars)	—	0.9s
edge	medium (~900 chars)	—	1.2s
edge	long (~2400 chars)	—	1.8s
elevenlabs	short	—	1.6s
elevenlabs	medium	—	8.6s
elevenlabs	long	—	20.1s
openai	short	—	4.9s
openai	medium	—	14.0s
openai	long	—	40.0s
inworld	short	—	3.3s
inworld	medium	—	5.3s
inworld	long	—	10.1s
inworld (progressive)	long	4.1s	8.8s

Edge is the fastest provider overall. For premium voices on longer texts, Inworld with progressive chunked playback starts audio 60% faster -- you hear the first chunk in ~4s instead of waiting ~10s for full synthesis.

How It Works

Text goes to edge-tts (or your chosen provider), comes back as MP3, and gets cached in ~/.speakly/cache/. The player window appears instantly in a loading state while TTS generates in the background. On a cache hit, playback starts immediately with no waiting.

Long text is automatically chunked and concatenated so there is no input length limit.

Contributing

git clone https://github.com/noncuro/speakly.git
cd speakly
uv sync
uv run speakly "test"

Fork it, branch it, send a PR. All contributions welcome.

License

MIT