The full directory.
Browse 50 models. Filter by license, VRAM, language, capability. Tap the checkbox on any card to add it to the comparison bar.
All 50 entries have a source-based license review. Some restrict commercial use or have unresolved weight terms. Each model page separates code, weights and additional conditions.
50 models · 29 with samples
Chatterbox
DefaultEnglish voice cloning with emotion-exaggeration controls
Added 2026-09-14 · Listen below
Chatterbox Nano
Bella referenceCompact English speech for CPU and on-device experiments.
Added 2026-09-14 · Listen below
Chatterbox Turbo
AndyEnglish voice cloning and expressive speech with a smaller model.
CosyVoice 2
DefaultAlibaba's real-time TTS, 150ms streaming latency
Ultra-realistic two-speaker dialogue, sounds like a podcast
F5-TTS
Reference cloneFlow-matching TTS, fast zero-shot voice cloning
Fish Speech 1.5
Reference cloneEarlier Fish Speech voice-cloning model; compare with S2 Pro
Higgs Audio v2
DefaultBuilt on Llama 3.2 3B, 10M+ hours of training audio
IndexTTS 2
DefaultBilibili's TTS with multi-speaker and precise emotion control
KittenTTS
Default15M params, under 25MB, runs anywhere — even your phone
Llasa
Bella referenceLLM-native TTS built straight on the Llama architecture
Added 2026-09-14 · Listen below
MOSS-TTS Nano
Bella referenceCompact CPU speech generation with streaming and voice cloning.
NeuTTS Air
Reference-basedOn-device-class TTS with instant cloning from ~3 seconds
OpenVoice v2
DefaultTone-color cloning + cross-lingual transfer
Llama-3 based, empathetic, low-latency for interactive apps
Describe the voice in natural language ('soft female, fast, clear')
Piper
LJ SpeechTiny, fast, runs offline on a Raspberry Pi
Added 2026-09-14 · Listen below
Pocket TTS
AlbaCPU speech generation, streaming, and voice cloning in a small model.
Qwen3-TTS
DefaultAlibaba's flagship, 97ms latency, 10 languages
Sesame CSM-1B
Context-conditionedThe conversational speech model behind the viral Maya demo
Spark-TTS
DefaultBuilt on Qwen2.5, zero-shot voice cloning + style control
StyleTTS 2
DefaultDiffusion-based, human-level naturalness on LibriTTS
VibeVoice
DefaultMicrosoft's long-form TTS — 90 minutes, 4 speakers
WhisperSpeech
DefaultWhisper, but inverted — TTS by 'unwrapping' OpenAI's ASR
XTTS v2
Reference cloneVoice cloning from a short reference clip across 17 languages
Zonos
Reference-based44kHz studio-quality output with emotion control
Added 2026-09-14 · Samples pending
Chatterbox Multilingual V3
DefaultCross-language voice cloning with the V3 multilingual checkpoint.
No samples yet
Added 2026-09-14 · Samples pending
CosyVoice 3
DefaultStreaming multilingual synthesis with pronunciation and instruction controls.
No samples yet
Added 2026-09-14 · Samples pending
Fish Audio S2 Pro
DefaultExpressive multilingual speech with inline emotion control and multi-speaker support.
No samples yet
Added 2026-09-14 · Samples pending
48 kHz stereo speech and cloning with the Local Transformer v1.5 architecture.
No samples yet
Added 2026-09-14 · Samples pending
MOSS-TTS v1.5
DefaultLong-form multilingual narration with cloning and explicit pause controls.
No samples yet
Muyan-TTS
Reference-basedFull training code released, trained for ~$50K on podcast audio
No samples yet
Step-Audio-EditX
Reference-basedRe-edit the emotion or style of audio you already generated
No samples yet
Added 2026-09-14 · Samples pending
TADA 3B Multilingual
DefaultAligned text and speech generation with multilingual reference prompts.
No samples yet
Added 2026-09-20 · Samples pending
VoxCPM2
Samples pendingMultilingual speech with voice design and reference conditioning
No samples yet
Voxtral TTS
Reference-basedMistral's zero-shot cloning model, ~70ms latency across 9 languages
No samples yet