TTS model pages

Local TTS models

Static, indexable pages for open local text-to-speech, voice cloning, ASR and speech AI models.

Interactive TTS list Run recommender

TTS / speech models

Featured

Best quality

Higgs Audio v2

Fastest

Parakeet TDT 0.6B v2

Featured TTS models

Alibaba Cloud (Qwen Team)

Qwen3 TTS

Quality 9.5 · Speed 8.5 · 2.8GB · Apache 2.0

State-of-the-art multilingual TTS with natural prosody and emotion control. Supports 30+ languages with streaming inference.

streamingrealtimemultilingualemotion

Rhasspy

Piper

Quality 7.5 · Speed 10 · 0.1GB · MIT

Fast, local neural TTS optimized for Raspberry Pi and edge devices. Lightweight with good quality for embedded systems.

realtimelow-latency

Speech Research

F5-TTS

Quality 9.4 · Speed 9 · 1.5GB · MIT

Flow-matching based TTS with SOTA quality and extremely fast inference. Simple and efficient architecture.

realtimecloningstreaming

hexgrad

Kokoro TTS

Quality 9.2 · Speed 9.8 · 0.33GB · Apache 2.0

Ultra-lightweight yet stunning quality. 82M params only - runs on CPU in real-time. Best quality-to-size ratio of any TTS model.

realtimestreaminglow-latencymultilingual

Canopy Labs

Orpheus TTS

Quality 9.6 · Speed 7.5 · 3.5GB · Apache 2.0

LLM-based TTS with human-level naturalness. Supports rich emotion tags (laugh, sigh, hesitation). Built on Llama 3 architecture for unprecedented expressiveness.

emotionstreamingcloning

Resemble AI

Chatterbox TTS

Quality 9.4 · Speed 8 · 1.2GB · MIT

Open-source SOTA voice cloning from Resemble AI. Outperforms ElevenLabs on naturalness benchmarks. Supports emotion exaggeration control and ultra-stable generation.

cloningemotionstreaming

Nari Labs

Dia

Quality 9.3 · Speed 7 · 3GB · Apache 2.0

1.6B dialogue TTS - generates realistic two-speaker conversations from a single transcript. Supports non-verbal cues like [laughs], [coughs], [sighs] natively.

dialogueemotioncloningstreaming

OuteAI

OuteTTS

Quality 8.7 · Speed 8.5 · 0.9GB · MIT

Pure language model approach to TTS - no separate audio encoder. Runs via llama.cpp for fully local GGUF inference. Excellent for CPU-only setups.

realtimelow-latencycloning

Alibaba FunAudioLLM

CosyVoice 2

Quality 9.3 · Speed 8.8 · 2.4GB · Apache 2.0

Industrial-grade multilingual TTS with streaming, voice cloning and emotion control. Exceptional Chinese + English quality. Used in production at Alibaba scale.

streamingrealtimecloningemotionmultilingual

SparkAudio

Spark TTS

Quality 9 · Speed 8.2 · 3GB · Apache 2.0

Bilingual TTS with virtual speaker creation - control pitch, speed, gender from text. Built on Qwen2.5 LLM backbone for powerful generation.

cloningstreamingrealtime

jamiepine / Community

Desktop app & orchestrator for local TTS - not a model. Provides a UI studio, voice profile management, and a local API. Generates audio via swappable backends (Qwen3 TTS, Kokoro, Piper, XTTS…). Think of it as a front-end shell that runs on top of your installed TTS models.

streamingrealtimelow-latency

Sesame AI

Sesame CSM

Quality 9.5 · Speed 7.5 · 3.5GB · Apache 2.0

Conversational Speech Model - generates speech with natural turn-taking, backchannels and interruptions. Built specifically for multi-turn dialogue with real-time response generation.

dialoguestreamingrealtimeemotion

All TTS model pages