Top TTS Fine-Tunes

Best fine-tuned text-to-speech models for voice generation and audio synthesis.

Last updated April 3, 2026 · Updated daily

Kokoro-82M by hexgrad holds the #1 position with 9.7M downloads, ahead of Kokoro-82M-bf16 at 775.0K.

The top 10 is dominated by hexgrad, mlx-community, microsoft. This is the first snapshot — future updates will track position changes and emerging trends.

The gap between #1 and #28 is 9.7M vs 3.8K downloads, showing massive concentration at the top.

🥇new

Kokoro-82M

Silver63

hexgrad · TTS · from yl4579/StyleTTS2-LJSpeech

9.7M

downloads

🥈new

Kokoro-82M-bf16

Bronze43

mlx-community · TTS · from yl4579/StyleTTS2-LJSpeech

775.0K

downloads

🥉new

VibeVoice-Realtime-0.5B

Silver51

microsoft · TTS · from Qwen/Qwen2.5-0.5B

447.0K

downloads

4new

Hypa_Orpheus-3b-0.1-ft-unsloth-merged_16bit

Bronze33

hypaai · TTS · from adapter:canopylabs/orpheus-3b-0.1-ft

167.7K

downloads

5new

orpheus-urdu-tts

Bronze35

mahwizzzz · TTS · from canopylabs/orpheus-3b-0.1-pretrained

122.1K

downloads

6new

tada-3b-ml

Bronze42

HumeAI · TTS · from meta-llama/Llama-3.2-3B

103.0K

downloads

7new

Kokoro-82M-v1.0-ONNX

Bronze42

onnx-community · TTS · from hexgrad/Kokoro-82M

93.2K

downloads

8new

MioTTS-2.6B

Bronze39

Aratako · TTS · from LiquidAI/LFM2-2.6B

79.4K

downloads

9new

parkiet

Bronze34

pevers · TTS · from nari-labs/Dia-1.6B

76.6K

downloads

10new

tada-1b

Bronze42

HumeAI · TTS · from meta-llama/Llama-3.2-1B

68.1K

downloads

11new

svara-tts-v1

Bronze35

kenpath · TTS · from adapter:canopylabs/3b-hi-ft-research_release

37.0K

downloads

12new

Kokoro-82M-ONNX

Bronze39

onnx-community · TTS · from hexgrad/Kokoro-82M

35.3K

downloads

13new

orpheus-3b-0.1-ft

Bronze32

unsloth · TTS · from canopylabs/orpheus-3b-0.1-ft

30.1K

downloads

14new

Kokoro-82M

Bronze25

Nextcloud-AI · TTS · from hexgrad/Kokoro-82M

28.8K

downloads

15new

mms-tts-eng

Bronze25

Xenova · TTS · from facebook/mms-tts-eng

26.2K

downloads

16new

F5-TTS_RUSSIAN

Bronze39

Misha24-10 · TTS · from SWivid/F5-TTS

22.6K

downloads

17new

react-native-executorch-kokoro

New

software-mansion · TTS · from hexgrad/Kokoro-82M

17.6K

downloads

18new

Llasa-1B

Bronze35

HKUSTAudio · TTS · from meta-llama/Llama-3.2-1B-Instruct

11.3K

downloads

19new

VieNeu-TTS

Bronze33

pnnbao-ump · TTS · from neuphonic/neutts-air

8.0K

downloads

20new

OmniVoice

Bronze34

k2-fsa · TTS · from Qwen/Qwen3-0.6B

6.6K

downloads

21new

s2-pro-gguf

Bronze30

rodrigomt · TTS · from fishaudio/s2-pro

5.8K

downloads

22new

orpheus-3b-0.1-ft-unsloth-bnb-4bit

Bronze29

unsloth · TTS · from canopylabs/orpheus-3b-0.1-ft

5.3K

downloads

23new

orpheus-3b-0.1-ft-Q4_K_M-GGUF

Bronze32

isaiahbjork · TTS · from canopylabs/orpheus-3b-0.1-ft

5.1K

downloads

24new

YarnGPT2

Bronze26

saheedniyi · TTS · from HuggingFaceTB/SmolLM2-360M

4.5K

downloads

25new

MIDI-LLM_Llama-3.2-1B

Bronze29

slseanwu · TTS · from meta-llama/Llama-3.2-1B

4.4K

downloads

26new

Qwen3-TTS-12Hz-1.7B-CustomVoice-8bit

Bronze29

mlx-community · TTS · from Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice

4.4K

downloads

27new

orpheus-3b-0.1-pretrained

Bronze34

canopylabs · TTS · from meta-llama/Llama-3.2-3B-Instruct

4.0K

downloads

28new

sesame-csm-elise

Bronze26

keanteng · TTS · from sesame/csm-1b

3.8K

downloads

About the Top TTS Fine-Tunes Leaderboard

Best fine-tuned text-to-speech models for voice generation and audio synthesis. This leaderboard tracks the top 28 fine-tuned models ranked by downloads, with daily snapshots to monitor how the rankings evolve over time.

Unlike HuggingFace's default model hub, Fine-Tune Catalog filters out pure quantizations and format conversions (repos that simply re-package a model as GGUF, AWQ, GPTQ, or EXL2 without any new training). Fine-tuned models published in quantized formats are still included — what matters is whether new training happened, not the output format.

Methodology

Rankings are based on total all-time download counts from HuggingFace. Downloads reflect real-world adoption — models and datasets that people actually use in production and research, not just stars or hype.

Rankings are snapshotted daily at 6:00 AM UTC. Position changes shown on the leaderboard compare the current snapshot to the previous day's snapshot. All data is sourced directly from the HuggingFace Hub API and processed through our classification pipeline, which uses tag analysis, model card parsing, and naming pattern detection to identify genuine fine-tunes.

Data Sources

  • HuggingFace Hub API — download counts, likes, trending scores, model metadata, and README/model cards
  • Model card parsing — training datasets, training method (LoRA, DPO, SFT, etc.), framework, hardware, and hyperparameters extracted from README files
  • Tag classification — fine-tune detection via `base_model:finetune:*` and `base_model:quantized:*` HuggingFace tags, plus naming pattern analysis

Who Is This For?

This leaderboard is designed for developers working in specific AI domains who need the best model for their particular use case — whether that's language understanding, computer vision, image generation, voice synthesis, or semantic search.

Whether you're a beginner exploring what's possible with fine-tuned AI models or an experienced ML engineer looking for the best starting point for your next project, these rankings give you a data-driven way to find the highest quality models without having to wade through thousands of quantizations, format conversions, and abandoned repositories on HuggingFace.

Update Schedule

This leaderboard was last updated on April 3, 2026. Rankings are refreshed daily with the latest download counts, likes, and trending data from HuggingFace. Historical snapshots are preserved to track trends over time — you can see which models are gaining traction and which are losing momentum.

Explore more rankings: Most Downloaded Fine-Tunes · Most Liked Fine-Tunes · Highest Quality Fine-Tunes · Top Uncensored Models · Top LLM Fine-Tunes · Top VLM Fine-Tunes and 16 more leaderboards