Backends

SciTeX Audio supports four TTS backends with automatic fallback. When no backend is specified, the system tries each in order until one succeeds.

Fallback Order

  1. ElevenLabs → 2. LuxTTS → 3. Google TTS → 4. System TTS

ElevenLabs

High-quality commercial TTS with natural-sounding voices.

  • Quality: Best

  • Cost: Paid (API key required)

  • Internet: Required

  • Default speed: 1.2x

speak("Hello", backend="elevenlabs", speed=1.2)

export SCITEX_AUDIO_ELEVENLABS_API_KEY="your-api-key"

LuxTTS

Open-source TTS with voice cloning support. Based on LuxTTS (Apache 2.0).

  • Quality: High (48kHz output)

  • Cost: Free

  • Internet: First model download only

  • Default speed: 2.0x

  • CPU performance: ~1.27x realtime factor (near-realtime)

speak("Hello", backend="luxtts", speed=2.0)

# Control CPU threads
speak("Hello", backend="luxtts", num_threads=4)

Features:

  • Voice cloning from reference audio samples

  • Singleton model caching (7.5s init, then instant)

  • Encoded prompt caching per reference audio

  • Automatic device detection (CUDA/MPS/CPU)

Place reference audio files in ~/.scitex/audio/reference/ for voice cloning. Without reference audio, a generic voice is used.

Google TTS (gTTS)

Free cloud-based TTS via Google Translate’s speech synthesis.

  • Quality: Good

  • Cost: Free

  • Internet: Required

  • Default speed: 1.5x

speak("Bonjour", backend="gtts", voice="fr", speed=1.5)

Supports many languages via the voice parameter (ISO language codes).

System TTS (pyttsx3)

Offline TTS using the system’s speech synthesis engine (espeak on Linux, SAPI5 on Windows).

  • Quality: Basic

  • Cost: Free

  • Internet: Not required

  • Default speed: 150 words per minute

speak("Hello", backend="pyttsx3", rate=200)

# Linux: install espeak
apt install espeak-ng

Checking Available Backends

from scitex_audio import available_backends, FALLBACK_ORDER

# List installed backends
print(available_backends())  # ['gtts', 'pyttsx3']

# See the fallback order
print(FALLBACK_ORDER)  # ['elevenlabs', 'luxtts', 'gtts', 'pyttsx3']

scitex-audio backends