Best text-to-speech models to run locally on a 12GB graphics card
A leaderboard of the best text-to-speech models you can run locally on a 12GB graphics card (with 32GB of system RAM). Voxtral TTS (4B) is #1 right now, followed by dots.tts MF 2B (MeanFlow, NFE 4) and FireRedTTS3 (base). Models are ranked by one score that combines 101 published benchmarks. Each model shows the download that fits and the command to run it.
109 models · scores combine 101 benchmarks · updated how it works · find a leaderboard · benchmark your machine
Models for 32GB RAM · 12GB graphics card → ~12GB fast · local only · reset
+ more
$ terminal
↗ share
detecting your OS…
Score vs memory to run
x
show
● open weights ◆ API only ─┘ best score for the memory · x: GB of RAM / VRAM for the smallest download (log)
| # | model | score↓ | where↓ | speed↓ | sources |
|---|---|---|---|---|---|
| 4 |  Voxtral TTS (4B) · bf16 | 71±4† | local |  | |
| 9 |  dots.tts MF 2B (MeanFlow, NFE 4) · bf16 (dots.tts package CLI) | 68±3† | local | | |
| 12 |  FireRedTTS3 (base) · bf16 (FireRedTTS3 repo) | 67±3 | local (slow) |  | |
| 16 |  LongCat-AudioDiT-3.5B · fp32 weights (official LongCat-AudioDiT repo, transformers>=5.3) | 66±2 | local (slow) |  | |
| 17 | LongCat-AudioDiT-1B · fp32 weights (official LongCat-AudioDiT repo, transformers>=5.3) | 66±4† | local | | |
| 18 |  OmniVoice · fp32 | 66±2 | local | | |
| 20 | dots.tts SOAR 2B · bf16 (dots.tts package CLI) | 65±2 | local | | |
| 23 |  Higgs Audio V3 TTS · bf16 | 65±3 | local | | |
| 25 |  Qwen3-TTS 12Hz 1.7B Base · bf16 (qwen-tts) | 65±1 | local | +3 | |
| 26 |  VoxCPM 0.5B · bf16 | 64±1 | local | | |
| 27 | VoxCPM2 · bf16 (voxcpm) | 64±1 | local | | |
| 28 |  Fun-CosyVoice3-0.5B-2512 · fp32 (CosyVoice repo) | 64±1 | local | | |
| 29 |  Wayu-Paxa-TTS-Edge · fp32 (wayu-tts package) | 64±4† | local | | |
| 32 |  IndexTTS-2 · fp32 (index-tts repo) | 63±1 | local | +2 | |
| 33 |  MOSS-TTS v1.5 (8B) · fp32 (transformers 5) | 63±2 | local (slow) | | |
| 34 |  GLM-TTS · fp32 (official GLM-TTS repo) | 63±2 | local | | |
| 37 |  Fish Audio S2 Pro · bf16 | 63±1 | local | +3 | |
| 39 | IndexTTS (v1) · fp32 on CPU (index-tts repo CLI `indextts`) | 62±3 | local | | |
| 40 |  F5-TTS v1 Base · fp32 | 62±3 | local | | |
| 42 | ZipVoice · fp32 (ZipVoice repo) | 62±3 | local | | |
| 44 | Qwen3-TTS 12Hz 0.6B Base · bf16 (qwen-tts) | 62±2 | local | | |
| 46 |  DMOSpeech 2 · fp32 (DMOSpeech2 repo) | 62±4† | local | | |
| 49 | CosyVoice2-0.5B · fp32 (CosyVoice repo) | 61±1 | local | +1 | |
| 50 | E2 TTS Base (F5-TTS reproduction) · fp32 | 60±2 | local | | |
| 53 |  MaskGCT · fp32 | 59±1 | local | +4 | |
| 60 | F5-TTS Base (v0) · fp32 | 57±2 | local | | |
| 63 |  ZONOS2 8B · q8_0 (zonos2.cpp) | 57±2† | local | | |
| 64 |  Supertonic 3 · onnx-fp32 (supertonic Python SDK 1.3.x) | 55±10* | local | | |
| 66 | Higgs Audio V2 · bf16 (transformers >=5.3.0) | 55±3 | local (slow) | | |
| 67 | Fasih-TTS-V1 · fp32 (coqui-tts 0.27.5) | 55±10* | local | | |
| 68 | CosyVoice-300M · fp32 (CosyVoice repo) | 54±3 | local | | |
| 69 |  Chatterbox Multilingual V3 · fp32 (chatterbox) | 54±10* | local | | |
| 70 |  VibeVoice 1.5B · bf16 (transformers >=5.17.0) | 53±6 | local | | |
| 73 |  Kokoro 82M v1.0 · fp32 (kokoro package 0.9.4) | 52±3 | local | | |
| 74 |  LEMAS-TTS (w/ prosody, multilingual_prosody) · official checkpoint (LEMAS-TTS repo, F5-TTS-based) | 52±6*† | local | | |
| 76 |  Spark-TTS 0.5B · fp32 (Spark-TTS repo CLI) | 52±3 | local | | |
| 77 |  Llasa-8B · bf16 (transformers + xcodec2) | 52±3 | local (slow) | | |
| 79 |  OpenVoice V1 · fp32 (myshell-ai/OpenVoice) | 51±14* | local | | |
| 84 | LEMAS-TTS (w/o prosody, multilingual_grl) · official checkpoint (LEMAS-TTS repo, F5-TTS-based) | 50±6*† | local | | |
| 85 |  MMS TTS Amharic · fp32 (transformers VitsModel) | 50±16* | local | | |
| 86 | Chatterbox · fp32 (chatterbox-tts 0.1.7) | 50±5 | local | | |
| 87 |  Parler-TTS Mini v0.1 · fp32 | 49±14* | local | | |
| 88 | Fish Speech 1.2 · bf16 (fish-speech repo) | 49±14* | local | | |
| 89 |  SILMA TTS v1 · fp32 checkpoint (silma-tts 1.0.5) | 48±10* | local | | |
| 90 | FireRedTTS (v1) · fp32 (FireRedTTS repo) | 48±3 | local | | |
| 92 |  Orpheus TTS 3B 0.1 (finetune-prod) · gguf Q4_K_M (orpheus-cpp 0.0.3) | 47±5 | local | | |
| 93 |  Kyutai TTS 1.6B en/fr · bf16 (moshi) | 47±5* | local | | |
| 95 | Amphion VALL-E (LibriLight 6k) · fp32 (Amphion repo recipe) | 47±14* | local | | |
| 96 |  Piper (en_US lessac medium) · onnx-fp32 (piper-tts 1.8.0) | 47±5* | local | | |
| 97 | Vevo (Amphion) · fp32 (Amphion repo) | 46±3 | local | | |
| 99 |  Dia2 2B · bf16 (nari-labs/dia2 repo CLI) | 45±6* | local | | |
| 102 | GPT-SoVITS v2 · fp32 (repo api_v2.py) | 44±4 | local | | |
| 103 | MeloTTS English · fp32 (MeloTTS repo CLI) | 44±5 | local | | |
| 104 | Parler-TTS Mini v1 · fp32 (parler-tts) | 44±6* | local | | |
| 106 | Nabra-82M v0.1 · fp32 (Kokoro KModel, Oddadmix/kokoro fork) | 42±10* | local | | |
| 107 | OpenAudio S1 Mini · bf16 (fish-speech official repo CLI) | 41±5 | local | | |
| 108 | Fish Speech 1.5 · bf16 (fish-speech repo) | 41±3 | local | | |
| 109 | EMA Lightning · fp32 (PyTorch) | 40±3* | local | | |
| 111 |  HierSpeech++ v1.1 · fp32 (repo inference.py) | 39±3 | local | | |
| 113 |  Magpie TTS Multilingual 357M · f16-gguf (NVIDIA NeMo-Speech.cpp) | 39±6* | local | | |
| 114 | Zonos v0.1 · bf16 | 39±4 | local | | |
| 115 |  Maya1 · bf16 | 38±6* | local | | |
| 117 |  Pheme 300M · fp32 (PolyAI-LDN/pheme repo) | 37±3 | local | | |
| 118 |  VoiceCraft 830M TTSEnhanced · fp32 (jasonppy/VoiceCraft repo) | 37±8* | local | | |
| 121 | Amphion VALL-E v2 recipe (MLS) · fp32 (Amphion repo) | 36±3 | local | | |
| 122 |  NeuTTS Air · bf16 (neutts 1.4.1) | 36±10* | local | | |
| 124 |  VITS (VCTK multi-speaker) · fp32 (jaywalnut310/vits repo) | 34±5 | local | | |
| 126 | VoiceCraft 830M (giga830M) · fp32 (VoiceCraft repo) | 32±4† | local | | |
| 127 |  VALL-E X (Plachtaa open implementation) · fp32 (Plachtaa/VALL-E-X repo) | 31±4 | local | | |
| 128 |  OuteTTS 0.2 500M · fp32 (outetts 0.3.3) | 31±6* | local | | |
| 130 | StyleTTS 2 (LibriTTS) · fp32 (styletts2 PyPI package 0.1.6) | 30±2 | local | +2 | |
| 131 |  XTTS v2 · fp32 (coqui-tts 0.27.5) | 30±2 | local | +7 | |
| 132 | Tortoise TTS v2 · fp16 (neonbjb/tortoise-tts) | 29±2 | local | +2 | |
| 133 | Parler-TTS Large v1 · fp32 | 29±3 | local | | |
| 134 |  WhisperSpeech v1.95 Medium (7-language) · fp16 (whisperspeech 0.8.9) | 28±3 | local | | |
| 136 | HierSpeech++ (LibriTTS-960) · fp32 (repo inference.py) | 28±5 | local | | |
| 137 | Piper (tr_TR dfki medium) · onnx-fp32 (piper-tts) | 26±4* | local | | |
| 138 | OpenVoice V2 · fp32 (myshell-ai/OpenVoice) | 26±3 | local | | |
| 139 |  Bark · fp32 (transformers) | 25±3 | local | | |
| 140 | SpeechT5 TTS · fp32 (transformers) | 23±3 | local | | |
| 141 |  MetaVoice 1B v0.1 · bf16 (metavoiceio/metavoice-src) | 20±3 | local | | |
| 143 | YourTTS · fp32 (coqui-tts) | 18±3 | local | | |
| 144 | MMS TTS Turkish · fp32 (transformers VitsModel) | 18±4* | local | | |
| 146 | Amphion NaturalSpeech 2 (LibriTTS) · fp32 (Amphion repo recipe) | 14±3 | local | | |
| 147 | Glow-TTS Common Voice (Turkish) · fp32 · CPU | 11±4* | local | | |
| — |  Breeze TTS 2 · bf16 | no reviewed results | local | ||
| — | Chatterbox Turbo · fp32 (chatterbox-tts 0.1.7) | no reviewed results | local | ||
| — | Dia 1.6B (0626) · fp32 (nari-labs/dia repo) | no reviewed results | local | ||
| — | FastPitch (Coqui LJ Speech) · fp32 · CPU | no reviewed results | local | ||
| — | FastPitch 1.1 + WaveGlow · fp32 (NVIDIA DeepLearningExamples FastPitch, NGC container) | not scored yet | local | ||
| — |  FastSpeech 2 + HiFi-GAN (SpeechBrain LJ Speech) · fp32 · CPU | no reviewed results | local | ||
| — | Glow-TTS (Coqui LJ Speech) · fp32 · CPU | no reviewed results | local | ||
| — | Glow-TTS Mai (Ukrainian) · fp32 · CPU | no reviewed results | local | ||
| — |  Grad-TTS + HiFi-GAN (10 steps) · fp32 (huawei-noah/Speech-Backbones Grad-TTS repo) | not yet calibrated | local | | |
| — |  Kitten TTS Mini 0.8 · onnx (kittenml 0.9.3) | no reviewed results | local | ||
| — | KittenTTS 2 · ternary-packed LM (kittenml 0.9.3) | no reviewed results | local | ||
| — | LightSpeech + Parallel WaveGAN · fp32 (microsoft/NeuralSpeech LightSpeech repo, user text) | no reviewed results | local | ||
| — | Neural HMM TTS (Coqui LJ Speech) · fp32 · CPU | no reviewed results | local | ||
| — | OverFlow (Coqui LJ Speech) · fp32 · CPU | no reviewed results | local | ||
| — | Pocket TTS (English) · fp32 (pocket-tts 3.3.0) | no reviewed results | local | ||
| — | PriorGrad acoustic + HiFi-GAN · fp32 (microsoft/NeuralSpeech PriorGrad-acoustic repo, T=12 fast sampling) | no reviewed results | local | ||
| — | SpeedySpeech (Coqui LJ Speech) · fp32 · CPU | no reviewed results | local | ||
| — | Tacotron 2 + WaveGlow (NVIDIA) · fp32 (torch.hub NVIDIA/DeepLearningExamples:torchhub) | not scored yet | local | ||
| — | Tacotron 2 DDC (Coqui LJ Speech) · fp32 · CPU | no reviewed results | local | ||
| — | VITS (LJ Speech) · fp32 (jaywalnut310/vits repo) | not yet calibrated | local | | |
| — | VITS Common Voice (Portuguese) · fp32 · CPU | no reviewed results | local | ||
| — | VITS CSS10 (French) · fp32 · CPU | no reviewed results | local | ||
| — | VITS CSS10 (Spanish) · fp32 · CPU | no reviewed results | local | ||
| — | VITS Thorsten (German) · fp32 · CPU | no reviewed results | local |
+ 78 more don't fit your machine or budget · show them
187 of 187 catalogue models listed here · sources and licences
* new or thinly tested: ranked on too few benchmark results · † only the maker’s own numbers so far · ~RAM is estimated; full context may need more memory