whatmodel.sh · find the best text-to-speech models you can run

Best text-to-speech models to run locally on a 4GB Mac

A leaderboard of the best text-to-speech models you can run locally on a Mac with 4GB of memory. VoxCPM2 is #1 right now, followed by Wayu-Paxa-TTS-Edge and IndexTTS (v1). Models are ranked by one score that combines 102 published benchmarks. Each model shows the download that fits and the command to run it.

38 models · 102 benchmarks · updated
Filter by your hardware and budget → ~3GB fast · local only · reset
ram[2][4][8][16][24][32][48][64][96][128][192][256]
[]
GB
gpu[apple silicon][graphics card][unified memory][none]
price[local][$1][$2][$5][$10][$20][$50][any]max per M chars
Filter by task
task[all tasks][voice cloning]
New #1s for your hardware:
orRSS

detecting your OS…

$ terminal

detecting your OS…

Run curl "whatmodel.sh/tts?ram=4&gpu=apple&max=0"to get this list in your terminal (add &format=json for agents).

Score vs release date

#modelscore↓where↓RAM↓speed↓get itsources
#modelscore↓where↓RAM↓speed↓get itsources
28VoxCPM2 · mlx-4bit (mlx-audio) OpenBMB64±1local~2.3GBhow to run › UltraEval-Audio (OpenBMB) · TTS replication resultsQwen-Audio-3.0-TTS technical reportZONOS2 Technical Report (ZTTS1-Eval results)nineninesix/gepard-1.0 model card (Seed-TTS-eval benchmark table)ZTTS1-Eval results snapshot (Zyphra)Arabic TTS Arena (Navid AI) · leaderboard snapshot 2026-09-24
29Wayu-Paxa-TTS-Edge · fp32 (wayu-tts package) Wayu Research / Paxa Labs64±4†local~0.3GBhow to run › Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech (Wayu Research / Paxa Labs / Typhoon; arXiv:2609.03502v1)
39IndexTTS (v1) · fp32 on CPU (index-tts repo CLI `indextts`) IndexTeam (bilibili)62±3local~1.2GBhow to run › DiSTAR: Diffusion over a Scalable Token Autoregressive Representation for Speech Generation (arXiv:2510.12210v2)
40F5-TTS v1 Base · fp32 SWivid (SJTU X-LANCE)62±3local~1.4GBhow to run › DiSTAR: Diffusion over a Scalable Token Autoregressive Representation for Speech Generation (arXiv:2510.12210v2)RVCBench: Benchmarking Robustness of Voice Cloning Across Modern Audio Generation Models (arXiv:2602.00443v2)
41ZipVoice · fp32 (ZipVoice repo) k2-fsa (Xiaomi)62±3local~0.6GBhow to run › OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models (arXiv:2604.00688v3)RVCBench: Benchmarking Robustness of Voice Cloning Across Modern Audio Generation Models (arXiv:2602.00443v2)
44Qwen3-TTS 12Hz 0.6B Base · bf16 (qwen-tts) Alibaba (Qwen)62±2local~2.5GBhow to run › UltraEval-Audio (OpenBMB) · TTS replication resultsQwen/Qwen3-TTS-12Hz-1.7B-Base model cardSGLang-Omni serving cookbooks (SGLang-Omni team measurements)
50E2 TTS Base (F5-TTS reproduction) · fp32 SWivid (SJTU X-LANCE)60±2local~1.3GBhow to run › DiSTAR: Diffusion over a Scalable Token Autoregressive Representation for Speech Generation (arXiv:2510.12210v2)F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching (arXiv:2410.06885v1)TTSDS2 listening test results (ttsds/listening_test)An Evaluation Framework for TTS Voice Reconstruction (SAP dysarthric speakers; arXiv:2606.21343v1)TTSDS2 multilingual benchmark results (ttsds/leaderboard_data)
60F5-TTS Base (v0) · fp32 SWivid (SJTU X-LANCE)57±2local~1.4GBhow to run › F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching (arXiv:2410.06885v1)TTSDS2 listening test results (ttsds/listening_test)TTSDS2 multilingual benchmark results (ttsds/leaderboard_data)
64Supertonic 3 · onnx-fp32 (supertonic Python SDK 1.3.x) Supertone55±10*local~0.4GBhow to run › Arabic TTS Arena (Navid AI) · leaderboard snapshot 2026-09-24
68CosyVoice-300M · fp32 (CosyVoice repo) Alibaba (FunAudioLLM)55±3local~2.3GBhow to run › F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching (arXiv:2410.06885v1)VERSA toolkit paper: TTS demonstration (arXiv:2412.17667v2)Amphion MaskGCT README · zero-shot TTS results table
73LEMAS-TTS (w/ prosody, multilingual_prosody) · official checkpoint (LEMAS-TTS repo, F5-TTS-based) LEMAS Project52±6*†local~1.5GBhow to run › LEMAS multilingual audio suite report
74Kokoro 82M v1.0 · bf16 hexgrad52±3local~0.7GBhow to run › EmergentTTS-Eval six-category leaderboardParadee: Distilling Kokoro-82M into an 8M-Parameter Single-Voice TTS Model (arXiv:2610.06817v1), Table 3 small-model comparisoncarun/tts-bench: intelligibility benchmark of four open TTS engines (published run, 2026-09-10)RW-Voice-EQ Bench (Hume AI)
79OpenVoice V1 · fp32 (myshell-ai/OpenVoice) MyShell51±15*local~0.3GBhow to run › TTSDS v1 published results
84LEMAS-TTS (w/o prosody, multilingual_grl) · official checkpoint (LEMAS-TTS repo, F5-TTS-based) LEMAS Project50±6*†local~1.4GBhow to run › LEMAS multilingual audio suite report
85MMS TTS Amharic · fp32 (transformers VitsModel) Meta50±16*local~0.2GBhow to run › Addis AI Amharic TTS benchmark
87Parler-TTS Mini v0.1 · fp32 Hugging Face (Parler-TTS)49±15*local~2.6GBhow to run › TTSDS v1 published results
90SILMA TTS v1 · fp32 checkpoint (silma-tts 1.0.5) SILMA AI48±10*local~2.7GBhow to run › Arabic TTS Arena (Navid AI) · leaderboard snapshot 2026-09-24
93Orpheus TTS 3B 0.1 (finetune-prod) · gguf Q4_K_M (orpheus-cpp 0.0.3) Canopy Labs47±5local~2.4GBhow to run › EmergentTTS-Eval six-category leaderboardEmergentTTS-Eval v1 paper (arXiv:2505.23009v1) · Table 1 automatic MOS
97Piper (en_US lessac medium) · onnx-fp32 (piper-tts 1.8.0) Open Home Foundation (Piper)47±5*local~0.1GBhow to run › Paradee: Distilling Kokoro-82M into an 8M-Parameter Single-Voice TTS Model (arXiv:2610.06817v1), Table 3 small-model comparison
103GPT-SoVITS v2 · fp32 (repo api_v2.py) RVC-Boss44±4local~1.1GBhow to run › TTSDS2 listening test results (ttsds/listening_test)TTSDS2 multilingual benchmark results (ttsds/leaderboard_data)
105MeloTTS English · fp32 (MeloTTS repo CLI) MyShell44±5local~0.7GBhow to run › VERSA toolkit paper: TTS demonstration (arXiv:2412.17667v2)Audio Deepfake Detection in the Age of Advanced Text-to-Speech models (arXiv:2601.20510v1)
107Nabra-82M v0.1 · fp32 (Kokoro KModel, Oddadmix/kokoro fork) oddadmix41±10*local~0.3GBhow to run › Arabic TTS Arena (Navid AI) · leaderboard snapshot 2026-09-24
108EMA Lightning · fp32 (PyTorch) EMA Lightning (canberkkkkkk)41±3*local~0GBhow to run › Antalia-2 Mini model card (cloud0day3/antalia-mini), public Freya-TR-Eval re-run tableEMA Lightning model card (canberkkkkkk/ema-lightning), Benchmarks: Freya-TR-Eval
110Fish Speech 1.5 · bf16 (fish-speech repo) Fish Audio40±3local~1.5GBhow to run › TTSDS2 listening test results (ttsds/listening_test)TTSDS2 multilingual benchmark results (ttsds/leaderboard_data)
114Magpie TTS Multilingual 357M · f16-gguf (NVIDIA NeMo-Speech.cpp) NVIDIA39±6*local~0.7GB~1×how to run › carun/tts-bench: intelligibility benchmark of four open TTS engines (published run, 2026-09-10)
128VALL-E X (Plachtaa open implementation) · fp32 (Plachtaa/VALL-E-X repo) Plachtaa (community)31±4local~1.6GBhow to run › VERSA toolkit paper: TTS demonstration (arXiv:2412.17667v2)ClonEval · Open Voice Cloning Leaderboard (AMU)CodecMOS-Accent: listening-test MOS of zero-shot TTS across English accents (arXiv:2603.14328v1)
129OuteTTS 0.2 500M · fp32 (outetts 0.3.3) OuteAI31±6*local~1.4GBhow to run › ClonEval · Open Voice Cloning Leaderboard (AMU)
131StyleTTS 2 (LibriTTS) · fp32 (styletts2 PyPI package 0.1.6) Yinghao Aaron Li et al. (Columbia University)30±2local~0.9GBhow to run › TTSDS2 listening test results (ttsds/listening_test)ZS-TTS-Evaluation (Edresson Casanova) · zero-shot TTS metrics of XTTS, YourTTS, TorToiSe and StyleTTS 2StyleTTS-ZS: Efficient High-Quality Zero-Shot Text-to-Speech Synthesis with Distilled Time-Varying Style Diffusion (arXiv:2409.10058v1)DiTTo-TTS: Diffusion Transformers for Scalable Text-to-Speech without Domain-Specific Factors (arXiv:2406.11427v2)An Evaluation Framework for TTS Voice Reconstruction (SAP dysarthric speakers; arXiv:2606.21343v1)RVCBench: Benchmarking Robustness of Voice Cloning Across Modern Audio Generation Models (arXiv:2602.00443v2)+2
132XTTS v2 · fp32 (coqui-tts 0.27.5) Coqui30±2local~1.9GBhow to run › Antalia-2 Mini model card (cloud0day3/antalia-mini), public Freya-TR-Eval re-run tableVoiceCraft: Zero-Shot Speech Editing and Text-to-Speech in the Wild (arXiv:2403.16973v3)ZS-TTS-Evaluation (Edresson Casanova) · zero-shot TTS metrics of XTTS, YourTTS, TorToiSe and StyleTTS 2StyleTTS-ZS: Efficient High-Quality Zero-Shot Text-to-Speech Synthesis with Distilled Time-Varying Style Diffusion (arXiv:2409.10058v1)DiTTo-TTS: Diffusion Transformers for Scalable Text-to-Speech without Domain-Specific Factors (arXiv:2406.11427v2)ClonEval · Open Voice Cloning Leaderboard (AMU)+7
138Piper (tr_TR dfki medium) · onnx-fp32 (piper-tts) Open Home Foundation (Piper)27±4*local~0.1GBhow to run › Antalia-2 Mini model card (cloud0day3/antalia-mini), public Freya-TR-Eval re-run table
139OpenVoice V2 · fp32 (myshell-ai/OpenVoice) MyShell25±3local~0.8GBhow to run › TTSDS2 listening test results (ttsds/listening_test)TTSDS2 multilingual benchmark results (ttsds/leaderboard_data)RVCBench: Benchmarking Robustness of Voice Cloning Across Modern Audio Generation Models (arXiv:2602.00443v2)TTSDS v1 published results
141SpeechT5 TTS · fp32 (transformers) Microsoft23±3local~0.7GBhow to run › TTSDS2 listening test results (ttsds/listening_test)TTSDS2 multilingual benchmark results (ttsds/leaderboard_data)
144MMS TTS Turkish · fp32 (transformers VitsModel) Meta18±4*local~0.2GBhow to run › Antalia-2 Mini model card (cloud0day3/antalia-mini), public Freya-TR-Eval re-run table
145YourTTS · fp32 (coqui-tts) Coqui / Edresson Casanova et al.18±3local~0.4GBhow to run › VoiceCraft: Zero-Shot Speech Editing and Text-to-Speech in the Wild (arXiv:2403.16973v3)ZS-TTS-Evaluation (Edresson Casanova) · zero-shot TTS metrics of XTTS, YourTTS, TorToiSe and StyleTTS 2DiTTo-TTS: Diffusion Transformers for Scalable Text-to-Speech without Domain-Specific Factors (arXiv:2406.11427v2)
—Kitten TTS Mini 0.8 · fp32 (mlx-audio) KittenMLno reviewed resultslocal~0.3GBhow to run ›
—KittenTTS 2 · ternary-packed LM (kittenml 0.9.3) KittenMLno reviewed resultslocal~2.1GBhow to run ›
—LightSpeech + Parallel WaveGAN · fp32 · CPU (microsoft/NeuralSpeech LightSpeech repo, LJSpeech test set) Microsoft NeuralSpeechno reviewed resultslocal~0GBhow to run ›
—Pocket TTS (English) · fp32 (pocket-tts 3.3.0) Kyutaino reviewed resultslocal~0.2GBhow to run ›

+ 151 more don't fit your machine or budget · show them

189 of 189 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

38 models · 102 benchmarks · updated