🔊 Ora

Ora is an open-source Linux alternative to Balabolka, built with GTK3 and Piper — 100% offline, neural text-to-speech.

---

Features

• Neural TTS via Piper — high-quality, fully offline after initial voice download
• 30+ languages (French, English, German, Spanish, and more)
• Voice auto-download — models fetched automatically on first use; no manual setup
• Sentence-level chunking — text is split into chunks for low-latency start and fine-grained progress reporting
• Chunk highlight — the currently playing sentence is highlighted in blue in the text area; the view auto-scrolls to follow along
• Per-chunk PCM cache with LRU eviction — instant re-playback without re-synthesis
• Pause / Resume — SIGSTOP/SIGCONT on the audio subprocess; no quality loss
• Restart — jump back to the first chunk without re-typing
• Clipboard watcher (opt-in) — auto-reads clipboard every 500 ms and speaks new content
• Clipboard quick-read button — paste and play in one click
• Progress bar — shows chunk-level progress (e.g. "Chunk 3/7")
• Voice quick-switch in the status bar — switch voices without opening Settings
• Settings dialog — manage installed voices, cache size, and clipboard watcher
• Offline mode — if no internet, only locally installed voices are available
• Speed control — ×0.5 to ×2 via a slider
• UI language auto-detected from system locale (French / English)
• All settings persisted across restarts: language, voice, speed, text, cache limits, clipboard preferences
• Dark theme

---

Requirements

System

sudo apt install python3-gi gir1.2-gtk-3.0 espeak-ng alsa-utils

Compatible with Ubuntu 22.04+ and Linux Mint 21+.

Piper

pip install piper-tts pathvalidate --break-system-packages

If pip is not available: sudo apt install python3-pip

pathvalidate is a dependency of piper-tts that is not always installed automatically — without it, no audio is produced.

---

Installation

git clone https://github.com/youruser/ora.git
cd ora
python3 -m ora

---

Usage

1. Select a voice from the status-bar dropdown — voices already downloaded are shown; others are downloaded automatically when you click ▶ Play.
2. Type or paste text in the text area.
3. ▶ Play — starts synthesis and playback, chunk by chunk.
4. ⏸ Pause / ▶ Resume — suspends or resumes audio at any point.
5. ⏮ Restart — restarts playback from the first chunk.
6. ⏹ Stop — interrupts playback immediately.
7. 📋 — reads clipboard content into the text area and starts playback.
8. ● (clipboard dot) in the status bar — click to toggle the automatic clipboard watcher.
9. ⚙ Settings — manage voices, cache, and clipboard preferences.

---

Data & config

All files are stored in ~/.local/share/ora/:

Path	Description
models/	Downloaded voice models (.onnx + .onnx.json)
cache/	Per-chunk PCM cache (one .raw file per synthesised chunk)
config.json	Persisted settings (see below)

config.json keys

Key	Type	Default	Description
lang	string	"fr"	Active voice language code
voice	string	""	Active voice key (e.g. fr_FR-tom-medium)
speed	float	1.0	Playback speed multiplier
text	string	""	Last text in the text area
cache_max_mb	int	200	Maximum PCM cache size in MB
clipboard_enabled	bool	false	Whether the clipboard watcher is active
clipboard_autostart	bool	false	Start the clipboard watcher on launch

---

Project structure

ora/
├── ora/                     # Python package (main application)
│   ├── __main__.py          # Entry point: python3 -m ora
│   ├── __init__.py
│   ├── app.py               # Main Gtk.Window — wires all modules together
│   ├── constants.py         # Paths, URLs, CSS, defaults
│   ├── config.py            # Config load/save (JSON persistence)
│   ├── i18n.py              # STRINGS dict + _() helper + locale detection
│   ├── core/
│   │   ├── tts.py           # TTSEngine: chunking, queue, synthesis via Piper
│   │   ├── cache.py         # CacheManager: LRU PCM cache
│   │   ├── clipboard.py     # ClipboardWatcher: GLib-based polling
│   │   └── voices.py        # VoiceManager: catalogue fetch, model download
│   └── ui/
│       └── settings_dialog.py  # SettingsDialog (Gtk.Dialog)
├── ora.py                   # Thin backward-compat launcher
├── Makefile                 # .deb packaging target
└── README.md

---

Recommended French voices

Voice	Gender	Quality
fr_FR-siwis-medium	Female	★★★★☆
fr_FR-mls-medium	Female	★★★☆☆
fr_FR-tom-medium	Male	★★★★☆

Models are downloaded from rhasspy/piper-voices on HuggingFace.

---

Building a .deb package

sudo apt install fakeroot dpkg-dev
make deb
sudo dpkg -i ora_1.0.0_all.deb

---

Roadmap

Items that are not yet implemented but planned:

Balabolka parity

• Open documents — load TXT, PDF, EPUB, DOCX, ODT, HTML and other formats directly into the text area
• Export to audio file — save the synthesised speech as MP3, WAV, OGG or FLAC
• Navigation controls — jump to previous / next sentence or paragraph during playback
• Bookmarks — save and restore named positions within a text

Lower priority

• Word-level highlight during playback — The Piper Python API supports phoneme-level timestamps (include_alignments=True) and word boundaries can be derived from the space phoneme in phoneme_alignments. However, this requires a voice model trained with duration output enabled. As of March 2026, no model in the official piper-voices catalogue has been trained with this option — every model returns a single ONNX output and phoneme_alignments is always None at runtime. The feature is blocked until new models are published.
• Flatpak packaging — needs a Flatpak manifest and GNOME runtime
• Systray support — would allow Ora to run in the background with a tray icon

---

License

GPL-3.0