local & free open-source voice-clone TTS models and music generation models.
Table of Contents
- Text-to-Speech (TTS) Models
- Music Generation Models
- Anything to Audio
- Audio Restoration & Enhancement
- Speech Recognition (ASR)
- Additional Resources
Text-to-Speech (TTS) Models
TTS Quick Comparison
| Model | Voice Cloning | ASR | Languages | Streaming | License |
|---|---|---|---|---|---|
| Voxtral-4B-TTS-2603 | ✔️ | ✔️ | 50+ | ✔️ | Apache-2.0 |
| LongCat-AudioDiT | ✔️ | ❌ | Zh/En | ❌ | MIT |
| Fish Audio S2 Pro | ✔️ | ❌ | 80+ | ✔️ | Research License |
| KittenTTS | ✔️ | ❌ | En+ | ✔️ | Apache-2.0 |
| MOSS-TTS | ✔️ | ❌ | 20 | ✔️ | Apache-2.0 |
| SoulX-Singer | ✔️ (Singing) | ❌ | Zh/En/Canto | ✔️ | Apache-2.0 |
| SoproTTS | ✔️ | ❌ | En | ✔️ | Apache-2.0 |
| NeuTTS | ✔️ | ❌ | En/Es/De/Fr | ✔️ | Apache-2.0 |
| Qwen3-TTS | ✔️ | ❌ | 10 | ✔️ | Apache-2.0 |
| GLM-TTS | ✔️ | ❌ | Zh/En | ✔️ | Apache-2.0 |
| VibeVoice-Realtime | ✔️ | ❌ | Multi | ✔️ | MIT |
| Fun-CosyVoice 3.0 | ✔️ | ❌ | 9 + 18 dialects | ✔️ | Apache-2.0 |
| MioTTS-2.6B | ✔️ | ❌ | En/Jp | ✔️ | LFM |
| Supertonic 2 | ❌ | ❌ | 5 | ✔️ | OpenRAIL-M |
| KugelAudio | ✔️ | ❌ | 23 EU | ✔️ | MIT |
| Kokoro-82M | ✔️ | ❌ | 8 (54 voices) | ✔️ | Apache-2.0 |
| KokoClone | ✔️ | ❌ | 7 | ✔️ | Apache-2.0 |
| IndexTTS2 | ✔️ | ❌ | Zh/En | ✔️ | Apache-2.0 |
| Maya1 | ✔️ | ❌ | En | ✔️ | Apache-2.0 |
| LFM2-Audio-1.5B | ✔️ | ✔️ | En | ✔️ | LFM |
| Step-Audio-EditX | ✔️ | ❌ | Zh/En/Jp/Ko | ✔️ | Apache-2.0 |
| FireRedTTS2 | ✔️ | ❌ | 7 langs | ✔️ | Apache-2.0 |
| VoxCPM | ✔️ | ❌ | Zh/En | ✔️ | Apache-2.0 |
| LuxTTS | ✔️ | ❌ | - | ✔️ | Apache-2.0 |
| MegaTTS3 | ✔️ | ❌ | Zh/En | ✔️ | Apache-2.0 |
| Spark-TTS | ✔️ | ❌ | Zh/En | ✔️ | Apache-2.0 |
| Fish Speech | ✔️ | ❌ | 8 langs | ✔️ | Apache-2.0 |
| Step-Audio | ✔️ | ✔️ | Zh/En/Jp | ✔️ | Apache-2.0 |
| SoulX-Podcast | ✔️ | ❌ | Zh/En/Canto | ✔️ | Apache-2.0 |
| Chatterbox | ✔️ | ❌ | 23+ | ✔️ | MIT |
| Orpheus-TTS | ✔️ | ❌ | Multi | ✔️ | Apache-2.0 |
| Dia | ✔️ | ❌ | En | ✔️ | Apache-2.0 |
| VieNeu-TTS | ✔️ | ❌ | Vi | ✔️ | Apache-2.0 |
| MiMo-Audio | ✔️ | ✔️ | Multi | ✔️ | Apache-2.0 |
| Kimi-Audio | ✔️ | ✔️ | Multi | ✔️ | MIT/Apache-2.0 |
| ZipVoice | ✔️ | ❌ | Zh/En | ✔️ | Apache-2.0 |
LongCat-AudioDiT
LongCat-AudioDiT
Description: State-of-the-art diffusion-based TTS model operating directly in waveform latent space. Developed by Meituan's LongCat team, it requires only a Waveform VAE and Diffusion backbone, effectively mitigating compounding errors.
Release Date: March 30, 2026
| Feature | Value |
|---|---|
| Parameters | 1B / 3.5B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ❌ |
| Emotion Control | ❌ |
| Languages | Chinese, English |
| Streaming | ❌ |
| Sample Rate | 24000 Hz |
| License | MIT |
Key Innovation: Adaptive Projection Guidance (APG) replaces traditional classifier-free guidance for elevated generation quality. Outperforms Seed-TTS on zero-shot voice cloning benchmarks.
Links:
Fish Audio S2 Pro
Fish Audio S2 Pro
Description: Fish Audio S2 Pro is a leading text-to-speech model with fine-grained inline control of prosody and emotion. It combines reinforcement learning alignment with a dual-autoregressive architecture for high-quality speech synthesis.
Release Date: March 10, 2026
| Feature | Value |
|---|---|
| Parameters | 5B (4B Slow AR + 400M Fast AR) |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ (15,000+ tags) |
| Emotion Control | ✔️ (fine-grained inline control) |
| Languages | 80+ (Tier 1: En, Zh, Jp) |
| Streaming | ✔️ (RTF 0.195, 100ms TTFA) |
| Model Size | ~10 GB (BF16) |
| License | Fish Audio Research License |
Links:
KittenTTS
KittenTTS
Description: KittenTTS is an open-source realistic text-to-speech model designed for lightweight deployment. It is a state-of-the-art TTS model under 25MB with just 15 million parameters, running without GPU on any device.
Release Date: February 24, 2026 (v0.8.1)
| Feature | Value |
|---|---|
| Parameters | 15M-80M |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ❌ |
| Emotion Control | ✔️ |
| Languages | English, Multiple |
| Streaming | ✔️ |
| License | Apache-2.0 |
Links:
MOSS-TTS
MOSS-TTS
Description: MOSS-TTS is a production-grade Text-to-Speech foundation model developed by OpenMOSS Team and MOSI.AI. Features state-of-the-art evaluation performance on Seed-TTS-eval benchmark with zero-shot voice cloning.
Release Date: February 10, 2026
| Feature | Value |
|---|---|
| Parameters | 8B (Delay), 1.7B (Local) |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ (Pinyin/Phoneme-level) |
| Emotion Control | ✔️ |
| Languages | 20 languages |
| Streaming | ✔️ |
| Max Duration | 1 hour |
| License | Apache-2.0 |
Links:
SoulX-Singer
SoulX-Singer
Description: SoulX-Singer is a high-fidelity, zero-shot singing voice synthesis model for generating realistic singing voices for unseen singers without fine-tuning.
Release Date: February 6, 2026
| Feature | Value |
|---|---|
| Parameters | - |
| Zero-shot Voice Cloning | ✔️ (Singing) |
| ASR | ❌ |
| Pronunciation Control | ✔️ (MIDI/F0) |
| Emotion Control | ✔️ |
| Languages | Mandarin, English, Cantonese |
| Streaming | ✔️ |
| License | Apache-2.0 |
Links:
SoproTTS
SoproTTS
Description: SoproTTS is a lightweight English text-to-speech model with zero-shot voice cloning. It uses dilated convolutions (WaveNet-style) and lightweight cross-attention layers instead of the common Transformer architecture.
Release Date: February 4, 2026 (v1.5)
| Feature | Value |
|---|---|
| Parameters | 135M |
| Zero-shot Voice Cloning | ✔️ (3-12s) |
| ASR | ❌ |
| Pronunciation Control | ❌ |
| Emotion Control | ✔️ (style_strength) |
| Languages | English |
| Streaming | ✔️ (250ms TTFA) |
| RTF | 0.05 (CPU M3) |
| Training Cost | ~$100 |
| License | Apache-2.0 |
Links:
NeuTTS
NeuTTS
Description: NeuTTS is a collection of open-source on-device TTS models with instant voice cloning. Built off LLM backbones with GGUF format quantizations for efficient on-device deployment.
Release Date: Early 2026
| Feature | Value |
|---|---|
| Parameters | 360M (Air), 120M (Nano) |
| Zero-shot Voice Cloning | ✔️ (3-second) |
| ASR | ❌ |
| Pronunciation Control | ❌ |
| Emotion Control | ❌ |
| Languages | English, Spanish, German, French |
| Streaming | ✔️ |
| On-Device | ✔️ (GGUF quantizations) |
| License | Apache-2.0 (Air), NeuTTS Open License 1.0 (Nano) |
Links:
Qwen3-TTS
Qwen3-TTS
Description: Qwen3-TTS is an open-source series of Text-to-Speech models developed by Alibaba Cloud. Supports stable, expressive, and streaming speech generation with free-form voice design.
Release Date: January 22, 2026
| Feature | Value |
|---|---|
| Parameters | 0.6B-1.7B |
| Zero-shot Voice Cloning | ✔️ (3-second) |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ |
| Languages | 10 (Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian) |
| Streaming | ✔️ (97ms latency) |
| License | Apache-2.0 |
Links:
GLM-TTS
GLM-TTS
Description: High-quality TTS synthesis system based on LLMs from ZhipuAI, supporting zero-shot voice cloning with Multi-Reward Reinforcement Learning.
Release Date: December 11, 2025
| Feature | Value |
|---|---|
| Parameters | - |
| Zero-shot Voice Cloning | ✔️ (3-10s) |
| ASR | ❌ |
| Pronunciation Control | ✔️ (Phoneme-level) |
| Emotion Control | ✔️ (RL-enhanced) |
| Languages | Chinese, English |
| Streaming | ✔️ |
| License | Apache-2.0 |
Links:
VibeVoice-Realtime
VibeVoice-Realtime
Description: Real-time TTS model from Microsoft with streaming text input and ultra-low latency (~300ms).
Release Date: December 3, 2025
| Feature | Value |
|---|---|
| Parameters | 0.5B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ |
| Languages | Multilingual |
| Streaming | ✔️ (300ms) |
| Max Duration | ~10 minutes |
| License | MIT |
Links:
Fun-CosyVoice 3.0
Fun-CosyVoice 3.0
Description: Advanced TTS system based on LLMs for zero-shot multilingual speech synthesis from FunAudioLLM.
Release Date: December 2025
| Feature | Value |
|---|---|
| Parameters | 0.5B |
| Zero-shot Voice Cloning | ✔️ (Multi-lingual/Cross-lingual) |
| ASR | ❌ |
| Pronunciation Control | ✔️ (Pinyin/CMU) |
| Emotion Control | ✔️ |
| Languages | 9 + 18+ Chinese dialects |
| Streaming | ✔️ (150ms) |
| License | Apache-2.0 |
Links:
MioTTS-2.6B
MioTTS-2.6B
Description: Lightweight, high-speed LLM-based TTS model for English and Japanese with minimal resource usage.
Release Date: 2026
| Feature | Value |
|---|---|
| Parameters | 2.6B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ❌ |
| Emotion Control | ❌ |
| Languages | English, Japanese |
| Streaming | ✔️ |
| RTF | 0.135-0.145 |
| License | LFM Open License |
Links:
Supertonic 2
Supertonic 2
Description: Lightning-fast, on-device text-to-speech system designed for extreme performance with minimal computational overhead. Powered by ONNX Runtime, it runs entirely on-device—no cloud, no API calls, no privacy concerns. Outperforms ElevenLabs Flash v2.5 by up to 42× in speed benchmarks.
Release Date: 2026
| Feature | Value |
|---|---|
| Parameters | 66M |
| Zero-shot Voice Cloning | ❌ |
| ASR | ❌ |
| Pronunciation Control | ❌ |
| Emotion Control | ❌ |
| Languages | English, Korean, Spanish, Portuguese, French |
| Streaming | ✔️ |
| RTF | 0.001-0.015 (up to 167× realtime) |
| On-Device | ✔️ (ONNX Runtime) |
| License | OpenRAIL-M |
Performance Comparison:
| System | Speed (chars/sec) | RTF |
|---|---|---|
| Supertonic 2 (RTX 4090) | 12,164 | 0.001 |
| Supertonic 2 (M4 Pro CPU) | 1,263 | 0.012 |
| ElevenLabs Flash v2.5 | 287 | 0.5 |
| Kokoro (Open-source) | 117 | 1.3 |
Links:
KugelAudio
KugelAudio
Description: Open-source TTS for European languages with 7B parameters. Outperformed ElevenLabs in human preference testing.
Release Date: Early 2026
| Feature | Value |
|---|---|
| Parameters | 7B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ (Speaking styles) |
| Languages | 23 European languages |
| Streaming | ✔️ |
| License | MIT |
Links:
Kokoro-82M
Kokoro-82M
Description: Kokoro is an open-weight Text-to-Speech model with 82 million parameters. Despite its lightweight architecture, it delivers comparable quality to larger models while being significantly faster and more cost-efficient. With Apache-licensed weights, Kokoro can be deployed anywhere from production environments to personal projects.
Release Date: January 27, 2025 (v1.0)
| Feature | Value |
|---|---|
| Parameters | 82M |
| Architecture | StyleTTS 2, ISTFTNet |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ (via misaki G2P) |
| Emotion Control | ✔️ (voice styles) |
| Languages | 8 (54 voices) |
| Streaming | ✔️ (generator pattern) |
| Cost | <$0.06 per hour of audio |
| License | Apache-2.0 |
Links:
KokoClone
KokoClone
Description: KokoClone is a fast, real-time compatible multilingual voice cloning system built on top of Kokoro-ONNX. It enables users to type text in multiple languages, provide a short 3-10 second reference audio clip, and instantly generate speech in that same voice.
Release Date: 2025
| Feature | Value |
|---|---|
| Parameters | 82M (Base: Kokoro-ONNX) |
| Zero-shot Voice Cloning | ✔️ (3-10s reference) |
| ASR | ❌ |
| Pronunciation Control | ❌ |
| Emotion Control | ✔️ |
| Languages | 7 (En, Hi, Fr, Ja, Zh, It, Pt, Es) |
| Streaming | ✔️ (CPU real-time) |
| License | Apache-2.0 |
Links:
IndexTTS2
IndexTTS2
Description: AI-Enhanced Text-to-Speech System with Intelligent Optimization and self-learning capabilities.
Release Date: November 2025
| Feature | Value |
|---|---|
| Parameters | - |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ (5 emotions) |
| Languages | Chinese, English |
| Streaming | ✔️ |
| Multi-speaker | ✔️ (1-4 speakers) |
| License | Apache-2.0 |
Links:
Maya1
Maya1
Description: State-of-the-art speech model for expressive voice generation with natural language voice control.
Release Date: November 2025
| Feature | Value |
|---|---|
| Parameters | 3B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ (Tags) |
| Languages | English (Multi-accent) |
| Streaming | ✔️ (<100ms) |
| License | Apache-2.0 |
Links:
LFM2-Audio-1.5B
LFM2-Audio-1.5B
Description: Liquid AI's first end-to-end audio foundation model with low latency and real-time conversation.
Release Date: November 28, 2025
| Feature | Value |
|---|---|
| Parameters | 1.5B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ✔️ (Integrated) |
| Pronunciation Control | N/A |
| Emotion Control | ✔️ |
| Languages | English |
| Streaming | ✔️ |
| License | LFM Open License |
Links:
Step-Audio-EditX
Step-Audio-EditX
Description: 3B-parameter LLM-based RL audio model specialized in expressive and iterative audio editing.
Release Date: November 2025
| Feature | Value |
|---|---|
| Parameters | 3B (4B BF16) |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ (Polyphone) |
| Emotion Control | ✔️ (14 emotions) |
| Languages | Mandarin, English, Sichuanese, Cantonese, Japanese, Korean |
| Streaming | ✔️ |
| License | Apache-2.0 |
Links:
FireRedTTS2
FireRedTTS2
Description: Long-form streaming TTS system for multi-speaker dialogue generation with stable, natural speech.
Release Date: September 2025
| Feature | Value |
|---|---|
| Parameters | - |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ |
| Languages | EN, ZH, JP, KO, FR, DE, RU |
| Streaming | ✔️ (140ms) |
| Multi-speaker | ✔️ (4 speakers) |
| Max Duration | 3 minutes |
| License | Apache-2.0 |
Links:
VoxCPM
VoxCPM
Description: Tokenizer-free TTS system for context-aware speech generation and true-to-life voice cloning.
Release Date: September 16, 2025
| Feature | Value |
|---|---|
| Parameters | 640M-800M |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ |
| Languages | Chinese, English |
| Streaming | ✔️ (RTF 0.17) |
| License | Apache-2.0 |
Links:
LuxTTS
LuxTTS
Description: Lightweight ZipVoice-based TTS model for high quality voice cloning at speeds exceeding 150x realtime.
Release Date: 2025
| Feature | Value |
|---|---|
| Parameters | - |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ❌ |
| Emotion Control | ❌ |
| Languages | - |
| Streaming | ✔️ |
| RTF | 150x |
| VRAM | 1GB |
| License | Apache-2.0 |
Links:
MegaTTS3
MegaTTS3
Description: Advanced zero-shot speech synthesis with Sparse Alignment Enhanced Latent Diffusion Transformer.
Release Date: March 22, 2025
| Feature | Value |
|---|---|
| Parameters | 0.45B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ |
| Languages | Chinese, English |
| Streaming | ✔️ |
| License | Apache-2.0 |
Links:
Spark-TTS
Spark-TTS
Description: Efficient LLM-Based TTS Model with Single-Stream Decoupled Speech Tokens, built on Qwen2.5.
Release Date: March 2025
| Feature | Value |
|---|---|
| Parameters | 0.5B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ |
| Languages | Chinese, English |
| Streaming | ✔️ |
| License | Apache-2.0 |
Links:
Fish Speech
Fish Speech
Description: State-of-the-art open source TTS and voice cloning model that generates natural, realistic, and emotionally rich speech.
Release Date: May 31, 2025 (v1.5.1)
| Feature | Value |
|---|---|
| Parameters | 4B (S1), 0.5B (S1-mini) |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ |
| Languages | 8 (EN, JP, KO, ZH, FR, DE, AR, ES) |
| Streaming | ✔️ |
| RTF | ~1:7 |
| License | Apache-2.0 |
Links:
Step-Audio
Step-Audio
Description: Production-ready open-source framework for intelligent speech interaction with unified speech comprehension and generation.
Release Date: February 17, 2025
| Feature | Value |
|---|---|
| Parameters | 130B (Chat), 3B (TTS) |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ✔️ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ |
| Languages | Chinese, English, Japanese |
| Streaming | ✔️ |
| License | Apache-2.0 |
Links:
Audio Flamingo 3 (AF3)
Audio Flamingo 3 (AF3)
Description: Fully open-source Large Audio Language Model from NVIDIA ADLR with state-of-the-art audio understanding.
Release Date: July 9, 2025
| Feature | Value |
|---|---|
| Parameters | 7B |
| Zero-shot Voice Cloning | ❌ |
| ASR | ✔️ |
| Pronunciation Control | N/A |
| Emotion Control | ✔️ |
| Languages | Multi-lingual |
| Streaming | ✔️ |
| Context | 10 minutes |
| License | Apache-2.0 |
Links:
SoulX-Podcast
SoulX-Podcast
Description: SOTA Multi-Speaker TTS model for generating realistic long-form podcasts with dialectal diversity.
Release Date: 2025
| Feature | Value |
|---|---|
| Parameters | - |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ |
| Languages | Mandarin, English, Cantonese, Sichuanese, Henanese |
| Streaming | ✔️ |
| Max Duration | 90+ minutes |
| License | Apache-2.0 |
Links:
Chatterbox
Chatterbox
Description: Family of SOTA open-source TTS models by Resemble AI with zero-shot voice cloning and multilingual synthesis.
Release Date: June 13, 2025 (v0.1.2)
| Feature | Value |
|---|---|
| Parameters | 350M-500M |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ (Tags) |
| Languages | 23+ |
| Streaming | ✔️ |
| License | MIT |
Links:
Orpheus-TTS
Orpheus-TTS
Description: SOTA open-source TTS built on Llama-3b backbone demonstrating emergent capabilities of LLMs for speech synthesis.
Release Date: April 2025
| Feature | Value |
|---|---|
| Parameters | 3B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ |
| Languages | Multilingual |
| Streaming | ✔️ (200ms) |
| License | Apache-2.0 |
Links:
Dia
Dia
Description: 1.6B parameter TTS model by Nari Labs for generating ultra-realistic dialogue in one pass.
Release Date: June 27, 2024
| Feature | Value |
|---|---|
| Parameters | 1.6B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ✔️ |
| Languages | English |
| Streaming | ✔️ |
| License | Apache-2.0 |
Links:
VieNeu-TTS
VieNeu-TTS
Description: Advanced on-device Vietnamese TTS model with instant voice cloning from 3-5 seconds of reference audio.
Release Date: 2025
| Feature | Value |
|---|---|
| Parameters | 0.3B-0.6B |
| Zero-shot Voice Cloning | ✔️ (3-5s) |
| ASR | ❌ |
| Pronunciation Control | ✔️ |
| Emotion Control | ❌ |
| Languages | Vietnamese |
| Streaming | ✔️ (On-device) |
| License | Apache-2.0 |
Links:
MiMo-Audio
MiMo-Audio
Description: Audio Language Model by Xiaomi functioning as a Few-Shot Learner with SOTA audio understanding.
Release Date: 2025
| Feature | Value |
|---|---|
| Parameters | 7B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ✔️ |
| Pronunciation Control | N/A |
| Emotion Control | ✔️ |
| Languages | Multi-lingual |
| Streaming | ✔️ |
| License | Apache-2.0 |
Links:
Kimi-Audio
Kimi-Audio
Description: Open-source audio foundation model by Moonshot AI for audio understanding, generation, and conversation.
Release Date: 2024
| Feature | Value |
|---|---|
| Parameters | 7B |
| Zero-shot Voice Cloning | ✔️ |
| ASR | ✔️ |
| Pronunciation Control | N/A |
| Emotion Control | ✔️ |
| Languages | Multi-lingual |
| Streaming | ✔️ |
| License | MIT/Apache-2.0 |
Links:
ZipVoice
ZipVoice
Description: Fast and high-quality zero-shot TTS models based on flow matching.
Release Date: June 16, 2025
| Feature | Value |
|---|---|
| Parameters | 123M |
| Zero-shot Cloning | ✔️ |
| Languages | Chinese, English |
| Dialogue | ✔️ |
| License | Apache-2.0 |
Links:
Music Generation Models
Music Quick Comparison
| Model | Music Gen | Languages | Streaming | License |
|---|---|---|---|---|
| ACE-Step 1.5 | ✔️ | 50+ | ✔️ | MIT |
| LeVo 2 | ✔️ | Zh/En | ❌ | Apache-2.0 |
| Foundation-1 | ✔️ (Samples) | - | ❌ | Stability AI |
| Music Flamingo | ❌ | - | - | Apache-2.0 |
| Magenta Realtime | ✔️ | - | ✔️ | Apache-2.0/CC-BY-4.0 |
| Uni-MoE (Audio) | ✔️ | - | ✔️ | Apache-2.0 |
ACE-Step 1.5
ACE-Step 1.5
Description: The most powerful local music generation model outperforming most commercial alternatives. Supports Mac, AMD, Intel, and CUDA devices.
Release Date: February 20, 2026 (v0.1.2)
| Feature | Value |
|---|---|
| Parameters | 0.6B-4B (LM), DiT variants |
| Music Generation | ✔️ |
| Lyrics Support | ✔️ (50+ languages) |
| Voice2BGM | ✔️ |
| Reference Audio | ✔️ |
| Track Separation | ✔️ |
| Duration | 10s - 10min |
| VRAM | <4GB |
| Platforms | CUDA, MPS, ROCm, XPU, CPU |
| License | MIT |
Links:
LeVo 2
LeVo 2 (SongGeneration 2)
Description: Open-source foundation model for commercial-grade music generation by Tencent AI Lab. It outperforms open-source baselines and rivals commercial systems in Overall Quality, Melody, Arrangement, Sound Quality, and Structure.
Release Date: 2025
| Feature | Value |
|---|---|
| Architecture | Hybrid LLM-Diffusion |
| Music Generation | ✔️ |
| Lyrics Support | ✔️ (Chinese, English) |
| Multilingual | ✔️ (Zh, En) |
| Text/Audio Prompts | ✔️ |
| VRAM | 12GB-22GB |
| License | Apache-2.0 |
Links:
Foundation-1
Foundation-1
Description: Structured text-to-sample generation model for music production workflows. Generates tempo-synced, key-aware, bar-aware sample generation with support for instrument identity, timbre control, and FX processing.
Release Date: 2025
| Feature | Value |
|---|---|
| Type | Text-to-Sample (Music) |
| Base Model | stabilityai/stable-audio-open-1.0 |
| Instrument Control | ✔️ |
| Timbre Descriptors | ✔️ (Warm, Bright, etc.) |
| FX Tags | ✔️ (Reverb, Delay, etc.) |
| Musical Notation | ✔️ (Chord, Melody, Arp) |
| VRAM | ~8GB |
| License | Stability AI Community License |
Links:
Music Flamingo
Music Flamingo
Description: Large audio-language model designed to advance music (including song) understanding. Achieves SOTA on 10+ music benchmarks.
Release Date: 2025
| Feature | Value |
|---|---|
| Parameters | - |
| Music Understanding | ✔️ |
| Music Generation | ❌ |
| Rich Captions | ✔️ |
| Music QA | ✔️ |
| Reasoning | ✔️ (Chain-of-thought) |
| Long-form | ✔️ |
| License | Apache-2.0 |
Links:
Magenta Realtime
Magenta Realtime
Description: Open music generation model from Google DeepMind enabling continuous generation of musical audio steered by text prompts or audio examples.
Release Date: August 2025
| Feature | Value |
|---|---|
| Parameters | - |
| Music Generation | ✔️ (Real-time) |
| Text-to-Music | ✔️ |
| Audio-to-Music | ✔️ |
| Reference Audio | ✔️ |
| Continuous Generation | ✔️ |
| Latency | Style prompt 2s+ |
| Context | 10 seconds |
| Training Data | ~190k hours |
| License | Apache-2.0 (code), CC-BY-4.0 (model) |
Links:
SoulX-Singer
SoulX-Singer
(Already listed in TTS - singing voice synthesis)
| Feature | Value |
|---|---|
| Parameters | - |
| Singing Generation | ✔️ |
| Zero-shot | ✔️ |
| Melody Control | ✔️ (F0/MIDI) |
| Languages | Mandarin, English, Cantonese |
| License | Apache-2.0 |
Uni-MoE (Audio)
Uni-MoE (Audio)
Description: MoE-based omnimodal model with voice cloning, TTS, T2M (text-to-music), and V2M (video-to-music).
Release Date: October 16, 2025 (Uni-MoE-Audio)
| Feature | Value |
|---|---|
| Parameters | - |
| Voice Cloning | ✔️ |
| TTS | ✔️ |
| Text-to-Music | ✔️ |
| Video-to-Music | ✔️ |
| Dynamic Routing | ✔️ |
| License | Apache-2.0 |
Links:
Anything to Audio
Models that can generate audio from multiple input modalities (video, text, image, audio). These are unified frameworks for multimodal audio synthesis.
Anything to Audio Quick Comparison
| Model | Text | Video | Image | Audio | License |
|---|---|---|---|---|---|
| PrismAudio | ❌ | ✔️ | ❌ | ❌ | Apache-2.0 |
| ThinkSound | ✔️ | ✔️ | ❌ | ✔️ | Apache-2.0 |
| HunyuanVideo-Foley | ✔️ | ✔️ | ❌ | ❌ | Research Only |
| MMAudio | ✔️ | ✔️ | ✔️ | ❌ | Apache-2.0 |
| AudioX | ✔️ | ✔️ | ✔️ | ✔️ | Apache-2.0 |
| Uni-MoE (Audio) | ✔️ | ✔️ | ❌ | ✔️ | Apache-2.0 |
AudioX
AudioX
Description: Unified framework for anything-to-audio generation integrating text, video, image, and audio conditions. Accepted to ICLR 2026.
Release Date: March 2025
| Feature | Value |
|---|---|
| Parameters | - |
| Text-to-Audio | ✔️ |
| Text-to-Music | ✔️ |
| Video-to-Audio | ✔️ |
| Audio Inpainting | ✔️ |
| Music Completion | ✔️ |
| License | Apache-2.0 |
Links:
MMAudio
MMAudio
Description: Multimodal joint training framework for high-quality synchronized audio generation from video and/or text inputs. State-of-the-art open source model for generating sounds for videos, images, and text prompts.
Release Date: December 2024 (CVPR 2025)
| Feature | Value |
|---|---|
| Parameters | - |
| Video-to-Audio | ✔️ |
| Text-to-Audio | ✔️ |
| Image-to-Audio | ✔️ |
| Synchronized Audio | ✔️ |
| Multimodal Joint Training | ✔️ |
| License | Apache-2.0 |
Links:
HunyuanVideo-Foley
HunyuanVideo-Foley
Description: Tencent's end-to-end video sound effect generation model for professional-grade AI Foley sound generation. Analyzes footage and creates immersive audio that matches the visual content perfectly.
Release Date: 2025
| Feature | Value |
|---|---|
| Parameters | - |
| Video-to-Audio (Foley) | ✔️ |
| Text-to-Audio | ✔️ |
| High-Quality Foley | ✔️ |
| Context-Aware | ✔️ |
| Output Quality | 48 kHz |
| License | Research & Non-commercial only |
Links:
ThinkSound
ThinkSound
Description: Unified Any2Audio generation framework with flow matching guided by Chain-of-Thought (CoT) reasoning. Supports generating or editing audio from video, text, audio, or their combinations. Accepted to NeurIPS 2025.
Release Date: 2025
| Feature | Value |
|---|---|
| Parameters | - |
| Video-to-Audio | ✔️ (SOTA) |
| Text-to-Audio | ✔️ |
| Audio-to-Audio | ✔️ |
| Audio Editing | ✔️ |
| CoT-Driven Reasoning | ✔️ |
| Interactive Object-centric Editing | ✔️ |
| License | Apache-2.0 (Research only) |
Links:
PrismAudio
PrismAudio
Description: Video-to-Audio generation framework with Reinforcement Learning and specialized Chain-of-Thought (CoT) planning. Decomposes reasoning into four specialized modules (Semantic, Temporal, Aesthetic, Spatial CoT) for comprehensive video understanding. Built upon ThinkSound.
Release Date: 2025 (ICLR 2026)
| Feature | Value |
|---|---|
| Parameters | 518M |
| Video-to-Audio | ✔️ |
| CoT Planning | ✔️ (4 modules) |
| Multi-Dimensional RL | ✔️ |
| Fast-GRPO | ✔️ (Hybrid ODE-SDE) |
| Inference Time | 0.63 seconds |
| License | Apache-2.0 |
Performance Benchmarks:
| Metric | VGGSound | AudioCanvas |
|---|---|---|
| Semantic (CLAP) | 0.47 | 0.52 |
| Temporal (DeSync↓) | 0.41 | 0.36 |
| Aesthetic (MOS-Q) | 4.21±0.35 | 4.12±0.28 |
Links:
Uni-MoE (Audio)
Uni-MoE (Audio)
Description: MoE-based omnimodal model with voice cloning, TTS, T2M (text-to-music), and V2M (video-to-music).
Release Date: October 16, 2025 (Uni-MoE-Audio)
| Feature | Value |
|---|---|
| Parameters | - |
| Voice Cloning | ✔️ |
| TTS | ✔️ |
| Text-to-Music | ✔️ |
| Video-to-Music | ✔️ |
| Dynamic Routing | ✔️ |
| License | Apache-2.0 |
Links:
Audio Restoration & Enhancement
Audio Restoration & Enhancement Quick Comparison
| Model | Type | Bandwidth Extension | Inpainting | License |
|---|---|---|---|---|
| NVIDIA A2SB | Restoration | ✔️ | ✔️ | NVIDIA Non-Commercial |
| NovaSR | Enhancement | ✔️ | ❌ | Apache-2.0 |
| AudioSR | Enhancement | ✔️ | ❌ | MIT |
NVIDIA A2SB
NVIDIA A2SB (Audio-to-Audio Schrodinger Bridges)
Description: Diffusion-based audio restoration model tailored for high-resolution music at 44.1kHz. An end-to-end, vocoder-free, multi-task model capable of both bandwidth extension (predicting high-frequency components) and inpainting (re-generating missing segments). Can restore hour-long audio inputs without boundary artifacts.
Release Date: January 2025
| Feature | Value |
|---|---|
| Architecture | End-to-end vocoder-free |
| Bandwidth Extension | ✔️ |
| Audio Inpainting | ✔️ |
| High-Resolution | ✔️ (44.1kHz) |
| Long Audio | ✔️ (hour-long) |
| Streaming | ❌ |
| License | NVIDIA OneWay NonCommercial License |
Links:
NovaSR
NovaSR
Description: Lightning fast audio upsampler - 50kB model that upscales 16kHz audio to 48kHz at 3500x realtime.
Release Date: 2025
| Feature | Value |
|---|---|
| Size | 52kB |
| Speed | 3600x realtime (A100) |
| Input | 16kHz |
| Output | 48kHz |
| VRAM | Minimal |
| License | Apache-2.0 |
Links:
AudioSR
AudioSR
Description: Audio super resolution model using latent diffusion to upscale low-quality audio to 48kHz.
Release Date: February 12, 2026 (v1.1.1)
| Feature | Value |
|---|---|
| Input | 8kHz-48kHz |
| Output | 48kHz |
| VRAM | 6GB min |
| Stereo | ✔️ |
| Long Audio | ✔️ |
| License | MIT |
Links:
Speech Recognition (ASR)
ASR Quick Comparison
| Model | Languages | Streaming | License |
|---|---|---|---|
| VibeVoice-ASR | 50+ | ✔️ | MIT |
| FunASR | 50+ | ✔️ | MIT |
VibeVoice-ASR
VibeVoice-ASR
Description: Microsoft's unified speech-to-text model for 60-minute long-form audio processing with speaker diarization and timestamping.
Release Date: January 21, 2026
| Feature | Value |
|---|---|
| Parameters | 7B |
| ASR | ✔️ |
| Languages | 50+ |
| Streaming | ✔️ |
| License | MIT |
Links:
FunASR
FunASR
Description: Fundamental end-to-end speech recognition toolkit with SOTA pretrained models.
Release Date: Ongoing (First: 2023)
| Feature | Value |
|---|---|
| ASR | ✔️ |
| VAD | ✔️ |
| Punctuation | ✔️ |
| Speaker Diarization | ✔️ |
| Multi-talker ASR | ✔️ |
| Timestamp | ✔️ |
| Emotion Recognition | ✔️ |
| Languages | 50+ |
| License | MIT/Model License |
Links:
Additional Resources
ComfyUI Integrations
ASR Leaderboard
| Resource | Description |
|---|---|
| Open ASR Leaderboard | Hugging Face leaderboard for comparing ASR model performance across languages and metrics. |
Links:
