Neura Market
Parler-TTS logo

Parler-TTS

Paid

Lightweight text-to-speech model for high-quality, controllable speech generation.

4.7
Text-to-SpeechContact
Inputs: textOutputs: audio
Type
Saas
Visit Website

About Parler-TTS

Parler-TTS is a fully open-source, lightweight text-to-speech (TTS) model developed by Hugging Face. It generates high-quality, natural sounding speech with controllable speaker attributes such as gender, pitch, and speaking style via a simple text prompt. The model is a reproduction of the paper 'Natural language guidance of high-fidelity text-to-speech with synthetic annotations' from Stability AI and Edinburgh University. Available in two sizes — Mini (880M parameters) and Large (2.3B parameters) — both trained on 45k hours of audiobook data. The repository provides both inference and training code, with optimizations like SDPA and Flash Attention 2 for faster generation. All datasets, preprocessing scripts, training code, and weights are released under a permissive license.

Key Features

High-quality, natural sounding speech generation with controllable speaker attributes (gender, pitch, speaking style)
Two model sizes: Mini (880M parameters) and Large (2.3B parameters), trained on 45k hours of audiobook data
Fully open-source: all datasets, preprocessing code, training code, and weights released under permissive license
Optimized for fast inference with SDPA and Flash Attention 2 compatibility and model compilation support
Simple text-based control via a prompt describing the desired speaker characteristics
Supports both random voice generation and specific speaker using pretrained checkpoints
Easy installation via pip from GitHub repository

Pros & Cons

Pros
  • Fully open-source with permissive license, encouraging community contributions and customization
  • High-quality speech output with fine-grained control over speaker attributes
  • Lightweight model design that runs efficiently on consumer hardware
  • Both inference and training code provided, enabling fine-tuning and customization
  • Two model sizes offer flexibility for different performance and quality needs
  • Advanced optimizations (SDPA, Flash Attention 2) for faster generation
Cons
  • Requires understanding of Python and machine learning frameworks to deploy and use effectively
  • Larger model (2.3B parameters) may need significant GPU memory for inference and training
  • As a research-oriented release, may lack polished user interface and documentation for non-technical users

Best For

Generating custom audiobook narrations with specific voice characteristicsCreating synthetic voiceovers for videos and presentationsBuilding interactive voice applications with controllable speech styleResearch and development of TTS models and voice cloningPrototyping and experimentation in text-to-speech applications

Alternatives to Parler-TTS

Readspeaker

Leading Voice Technology for 25 Years

AssemblyAI

Speech-to-text and NLP API

SpeechEasy

Effortless Synthetic Voice Solutions with Speecheasy™

Beepbooply

Transform Text into Natural-Sounding Speech with beepbooply's AI Voices

Deepgram

AI speech recognition API

AI Reads

Stay Informed with AI Reads: News Summarized and Read Aloud Anywhere

FAQ

What is Parler-TTS?
Parler-TTS is an open-source, lightweight text-to-speech model that generates high-quality, natural sounding speech with controllable speaker attributes like gender, pitch, and speaking style.
How do I install Parler-TTS?
Parler-TTS can be installed via pip: pip install git+https://github.com/huggingface/parler-tts.git. Apple Silicon users may need additional steps for bfloat16 support.
What model sizes are available?
Two checkpoints are available: Parler-TTS Mini (880M parameters) and Parler-TTS Large (2.3B parameters), both trained on 45k hours of audiobook data.
Can I use Parler-TTS for commercial projects?
Yes, all datasets, preprocessing code, training code, and weights are released under a permissive license, allowing commercial use.
How can I control the speaker style?
You control the speaker style by providing a text description (e.g., 'A female speaker delivers a slightly expressive and animated speech with a moderate speed and pitch') along with the text prompt.