Tortoise TTS
UltraUltra-High Quality Speech with Unmatched Naturalness
Very Slow
Speed
Exceptional
Quality
Yes
Cloning
1
Languages
About Tortoise TTS
Tortoise TTS is an autoregressive text-to-speech model that prioritizes audio quality above all else. Using a combination of autoregressive transformers and diffusion models, Tortoise generates extremely natural speech that captures subtle nuances of human voice. While slower than other models, Tortoise produces the most natural-sounding TTS output available.
Key Features
Ultra-High Quality
The most natural-sounding TTS output available.
Voice Cloning
Clone voices with exceptional fidelity and nuance.
Natural Prosody
Captures subtle speech patterns and micro-expressions.
Quality Presets
Choose from ultra_fast to high_quality processing.
Emotional Depth
Generates speech with genuine emotional resonance.
Open Source
Apache 2.0 licensed with commercial use rights.
Use Cases
Premium Audiobooks
Film Production
Documentary Narration
Professional Voiceovers
Archival Projects
High-End Content
Tortoise TTS Voices
View All 18Tortoise Angie
ENTortoise Deniro
ENTortoise Freeman
ENTortoise Geralt
ENTortoise Halle
ENTortoise Jlaw
ENTortoise Lj
ENTortoise Mol
ENTortoise Myself
ENTortoise Pat
ENTortoise Pat2
ENTortoise Snakes
ENFrequently Asked Questions
Tortoise TTS is an autoregressive text-to-speech model created by James Betker that prioritizes audio quality. It uses transformers and diffusion models to generate speech with unmatched naturalness and emotional depth.
Tortoise is open-source under Apache 2.0 license. On TextToSpeechAI, we charge 50 credits per 1000 characters (Ultra tier) due to extensive compute requirements and exceptional output quality.
Tortoise primarily supports English. It was trained on English speech datasets. For multilingual needs with similar quality, consider F5-TTS or use Tortoise in combination with other models.
Tortoise is the slowest TTS model due to its quality-first architecture. Generation can take 30 seconds to several minutes depending on text length and quality preset. Use "fast" preset for reasonable wait times.
Tortoise offers 4 presets: ultra_fast (testing), fast (production default), standard (balanced), and high_quality (maximum quality). Higher quality presets generate multiple candidates and select the best.
Provide multiple reference audio samples (ideally 3-10 clips, 5-10 seconds each) of the voice to clone. Tortoise analyzes these to capture voice characteristics, speaking patterns, and subtle nuances.
Tortoise produces exceptional audio quality - widely considered the most natural-sounding TTS available. It captures micro-expressions, breathing patterns, and emotional nuances that other models miss.
Tortoise requires 12-24GB of VRAM depending on quality preset and model size. High-end GPUs like RTX 3090, 4090, or A100 are recommended. CPU inference is possible but extremely slow.
Yes, Tortoise is Apache 2.0 licensed which permits commercial use with attribution. It is ideal for premium content where exceptional quality justifies longer processing times.
Select a Tortoise voice and optionally specify a quality preset in your API request. Note that generation times are longer than other models. We recommend the "fast" preset for most use cases.
Tortoise outputs high-quality WAV audio at 24kHz. Through TextToSpeechAI, you can request MP3, WAV, or OGG with quality-preserving encoding.
Tortoise produces the highest quality speech but is by far the slowest. Use it when quality is paramount and time is not a constraint. For faster results, StyleTTS 2 offers excellent quality. For real-time needs, use Piper.
Technical Specs
- Generation Speed Very Slow
- Output Quality Exceptional
- Voice Cloning Supported
- Languages 1
- GPU VRAM 12-24GB
- Credits/1000 chars 50