Kokoro-82M
A compact 82M text-to-speech model suited to local audio pipelines, CPU inference, and lightweight browser prototypes.
The brutal truth
Runs on CPU, tiny footprint, Apache 2.0 — the only reason not to use it is voice quality. If you need premium voices, pay for ElevenLabs.
Files and Versions
Direct inspected tree artifacts from Hugging Face. Verified filenames, exact byte sizes, model weight shards, and configuration definitions.
Files and versions are hosted on Hugging Face. You can view all branches, commits, model weights, and tokenizer artifacts directly.
Inspect hexgrad/Kokoro-82M /tree/main ↗Inspect /tree/main Repository ↗Decision matrix
✓ RUN / USE THIS IF…
- •You want CPU-only TTS
- •You need Apache 2.0
- •You're prototyping quickly
✕ WALK AWAY IF…
- •You need premium voice quality
- •You need voice cloning
- •You need 30+ languages
True cost of ownership
About
A compact 82M text-to-speech model suited to local audio pipelines, CPU inference, and lightweight browser prototypes.
What it does
Download the weights from Hugging Face, pick a quantisation that matches your GPU, and run it locally with Ollama, llama.cpp, vLLM, or a compatible runtime. Text-to-speech runs on your own hardware — no per-token bill, no third-party data handling.
Key features
Very small footprint for a TTS model
Runs on CPU without a discrete GPU
Use cases
Turn scripts into natural-sounding voiceover at scale.
Generate consistent show intros and transitions.
Prototype NPC voice lines before casting actors.
Convert articles and docs into listenable audio.
Quantization & Hardware Fit
VRAM Hardware Compatibility & Memory Allocation
| Quant | VRAM | Quality / Fit on 24GB | Install / Run Command (OLLAMA) | |
|---|---|---|---|---|
| ONNX CPU● ACTIVE | 2 GB | Standard · ✓ Fits (22GB spare) | python -m kokoro --text "Hello world" | |
| FP16 GPU | 4 GB | Standard · ✓ Fits (20GB spare) | python -m kokoro --text "Hello world" |
Buyer doubts, cleared
Can I actually run Kokoro-82M on my 24GB GPU?
What's the fine print on the license?
Does the Text-to-speech actually hold up in practice?
User reviews
Clean interface, consistent results once you land on a prompt you like. Wish there were more export presets for client handoffs.
Reasoning quality is close to the bigger models for structured tasks. Open-ended planning still needs a frontier model.
Switched our internal workflow to Kokoro-82M after the last pricing change. Nobody noticed a quality difference, and the bill dropped by half.
Kokoro-82M handled our batch jobs without surprises. The API is well documented and the rate limits are reasonable.
