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

Best voice cloning models to run locally on a 64GB Mac

A leaderboard of the best voice cloning models you can run locally on a Mac with 64GB of memory. LongCat-AudioDiT-3.5B is #1 right now, followed by VoxCPM2 and LongCat-AudioDiT-1B. Models are ranked on how clear the cloned voice is and how much it sounds like the original speaker. Each model shows the download that fits and the command to run it.

26 models · scores combine 13 benchmarks · updated how it works · find a leaderboard · benchmark your machine
task[all tasks][voice cloning]

Scores use reviewed voice cloning evaluations only. Compare scores within this task. Models without task measurements are excluded.

Models for 64GB RAM · Apple Silicon → ~48GB fast · local only · reset
+ more
chip[M1 Max][M1 Ultra][M2 Max][M2 Ultra][M3 Max (16-core CPU / 40-core GPU)][M4 Pro][M4 Max (16-core CPU / 40-core GPU)][M5 Pro][M5 Max (40-core GPU)]
set[all sets][Seed-TTS test-en · Whisper large-v3 WER · 43][MiniMax TTS · 9-language speaker similarity · 3]

Filter the model list while keeping task scores. Models without a shared score scale are not yet calibrated; select a set to see its own scores.

Detect your exact chip or graphics card:

curl -G whatmodel.sh/tts -d max=0 --data-urlencode "mac=$(sysctl -n hw.memsize machdep.cpu.brand_string)"Mac

curl -G whatmodel.sh/tts -d max=0 --data-urlencode "ram=$(awk '/MemT/{print $2}' /proc/meminfo)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>/dev/null)"Linux

curl.exe -G whatmodel.sh/tts -d max=0 --data-urlencode "ram=$((gcim Win32_ComputerSystem).TotalPhysicalMemory)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>$null)"Windows

$ terminal

detecting your OS…

Run curl "whatmodel.sh/tts?ram=64&gpu=apple&max=0"to get this list in your terminal (add &format=json for agents).

ram[8][16][24][32][48][64][96][128][192][256]GB
gpu[apple silicon][graphics card][none]
price[local][$20][$50][any]max per M chars

detecting your OS…

Score vs memory to run

#modeltask score↓where↓speed↓RAM↓get itsources
#modeltask score↓where↓speed↓RAM↓get itsources
4LongCat-AudioDiT-3.5B · fp32 weights (official LongCat-AudioDiT repo, transformers>=5.3) Meituan66±4local~15GBhow to run › Qwen-Audio-3.0-TTS technical reportmeituan-longcat/LongCat-AudioDiT-3.5B model card (Seed benchmark table)
8VoxCPM2 · bf16 (voxcpm) OpenBMB61±3local~5GBhow to run › UltraEval-Audio (OpenBMB) · TTS replication resultsQwen-Audio-3.0-TTS technical reportnineninesix/gepard-1.0 model card (Seed-TTS-eval benchmark table)
9LongCat-AudioDiT-1B · fp32 weights (official LongCat-AudioDiT repo, transformers>=5.3) Meituan60±6†local~5.7GBhow to run › meituan-longcat/LongCat-AudioDiT-3.5B model card (Seed benchmark table)
10Higgs Audio V3 TTS · bf16 Boson AI59±6*†local~10GBhow to run › Higgs Audio v3 comparison tables in SGLang-Omni cookbook (Boson AI contribution)
11OmniVoice · fp32 k2-fsa (Xiaomi)59±3local~3.3GBhow to run › Higgs Audio v3 comparison tables in SGLang-Omni cookbook (Boson AI contribution)OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models (arXiv:2604.00688v3)nineninesix/gepard-1.0 model card (Seed-TTS-eval benchmark table)nsanku TTS Benchmark (GhanaNLP community Space)
13Qwen3-TTS 12Hz 1.7B Base · bf16 (qwen-tts) Alibaba (Qwen)58±3local~4.5GBhow to run › UltraEval-Audio (OpenBMB) · TTS replication resultsHiggs Audio v3 comparison tables in SGLang-Omni cookbook (Boson AI contribution)Raon-OpenTTS: Open Models and Data for Robust Text-to-Speech (arXiv:2605.20830v2)Qwen/Qwen3-TTS-12Hz-1.7B-Base model card
14IndexTTS (v1) · fp32 on CPU (index-tts repo CLI `indextts`) IndexTeam (bilibili)57±7local~1.2GBhow to run › DiSTAR: Diffusion over a Scalable Token Autoregressive Representation for Speech Generation (arXiv:2510.12210v2)
16F5-TTS v1 Base · fp32 SWivid (SJTU X-LANCE)56±6local~1.4GBhow to run › DiSTAR: Diffusion over a Scalable Token Autoregressive Representation for Speech Generation (arXiv:2510.12210v2)
17Fish Audio S2 Pro · bf16 (fish-speech official repo CLI) Fish Audio56±3local~11GBhow to run › UltraEval-Audio (OpenBMB) · TTS replication resultsHiggs Audio v3 comparison tables in SGLang-Omni cookbook (Boson AI contribution)nineninesix/gepard-1.0 model card (Seed-TTS-eval benchmark table)Fish Audio S2 Technical Report
18Qwen3-TTS 12Hz 0.6B Base · bf16 (qwen-tts) Alibaba (Qwen)56±4local~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)
19IndexTTS-2 · fp32 (index-tts repo) IndexTeam (bilibili)54±3local~8.9GBhow to run › UltraEval-Audio (OpenBMB) · TTS replication resultsHiggs Audio v3 comparison tables in SGLang-Omni cookbook (Boson AI contribution)Raon-OpenTTS: Open Models and Data for Robust Text-to-Speech (arXiv:2605.20830v2)OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models (arXiv:2604.00688v3)
20ZipVoice · fp32 (ZipVoice repo) k2-fsa (Xiaomi)54±6local~0.6GBhow to run › OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models (arXiv:2604.00688v3)
22Fun-CosyVoice3-0.5B-2512 · fp32 (CosyVoice repo) Alibaba (FunAudioLLM)53±3local~5.4GBhow to run › UltraEval-Audio (OpenBMB) · TTS replication resultsRaon-OpenTTS: Open Models and Data for Robust Text-to-Speech (arXiv:2605.20830v2)
23E2 TTS Base (F5-TTS reproduction) · fp32 SWivid (SJTU X-LANCE)53±4local~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)
26MOSS-TTS v1.5 (8B) · fp32 (transformers 5) OpenMOSS53±3local~24GBhow to run › UltraEval-Audio (OpenBMB) · TTS replication resultsHiggs Audio v3 comparison tables in SGLang-Omni cookbook (Boson AI contribution)SGLang-Omni serving cookbooks (SGLang-Omni team measurements)
28F5-TTS Base (v0) · fp32 SWivid (SJTU X-LANCE)52±6†local~1.4GBhow to run › F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching (arXiv:2410.06885v1)
30Higgs Audio V2 · mlx-8bit (mlx-audio) Boson AI51±6*†local~7GBhow to run › Higgs Audio v3 comparison tables in SGLang-Omni cookbook (Boson AI contribution)
33CosyVoice2-0.5B · fp32 (CosyVoice repo) Alibaba (FunAudioLLM)48±3local~4.1GBhow to run › UltraEval-Audio (OpenBMB) · TTS replication resultsRaon-OpenTTS: Open Models and Data for Robust Text-to-Speech (arXiv:2605.20830v2)DMOSpeech 2: Reinforcement Learning for Duration Prediction in Metric-Optimized Speech Synthesis (arXiv:2507.14988v1)DiSTAR: Diffusion over a Scalable Token Autoregressive Representation for Speech Generation (arXiv:2510.12210v2)
35Spark-TTS 0.5B · fp32 (Spark-TTS repo CLI) SparkAudio41±5local~3.9GBhow to run › DMOSpeech 2: Reinforcement Learning for Duration Prediction in Metric-Optimized Speech Synthesis (arXiv:2507.14988v1)
36CosyVoice-300M · fp32 (CosyVoice repo) Alibaba (FunAudioLLM)40±5local~2.3GBhow to run › F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching (arXiv:2410.06885v1)Amphion MaskGCT README · zero-shot TTS results table
40XTTS v2 · fp32 (coqui-tts 0.27.5) Coqui23±6local~1.9GBhow to run › ZS-TTS-Evaluation (Edresson Casanova) · zero-shot TTS metrics of XTTS, YourTTS, TorToiSe and StyleTTS 2Amphion MaskGCT README · zero-shot TTS results table
41StyleTTS 2 (LibriTTS) · fp32 (styletts2 PyPI package 0.1.6) Yinghao Aaron Li et al. (Columbia University)17±7*local~0.9GBhow to run › ZS-TTS-Evaluation (Edresson Casanova) · zero-shot TTS metrics of XTTS, YourTTS, TorToiSe and StyleTTS 2
42Tortoise TTS v2 · fp16 (neonbjb/tortoise-tts) Tortoise12±7*local~5.6GBhow to run › ZS-TTS-Evaluation (Edresson Casanova) · zero-shot TTS metrics of XTTS, YourTTS, TorToiSe and StyleTTS 2
43YourTTS · fp32 (coqui-tts) Coqui / Edresson Casanova et al.-1±7*†local~0.4GBhow to run › ZS-TTS-Evaluation (Edresson Casanova) · zero-shot TTS metrics of XTTS, YourTTS, TorToiSe and StyleTTS 2
—Voxtral TTS (4B) · bf16 Mistralnot yet calibratedlocal~10GBhow to run › Voxtral TTS
—ZONOS2 8B · mlx-bf16 (mlx-audio) Zyphranot scored yetlocal~16GBhow to run ›

+ 21 more don't fit your machine or budget · show them

47 of 47 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