Kokoro vs Chatterbox

Kokoro is a tiny fixed-voice model. Chatterbox (Resemble AI, MIT) adds zero-shot voice cloning at 350M (Turbo, English) or 500M (Multilingual) parameters. We haven't benchmarked Chatterbox yet.

Side by side

Kokoro 82MChatterbox-TurboChatterbox (Multilingual)
OrganizationhexgradResemble AIResemble AI
Parameters82M350M500M
Licenseapache-2.0 sourcemit sourcemit source
Commercial useYes
Apache-2.0; card welcomes production/commercial deployment
Yes
MIT; watermarked outputs
Yes
MIT; outputs carry Perth neural watermark
Stated memoryNot statedNot statedNot stated
Runs on CPUYesUnknownUnknown
StreamingUnknownUnknownUnknown
Voice cloningUnknownYesYes
Languages9 languagesEnglish23+ languages
Our fastest CPU resultRTF 0.1234
kokoro-onnx fp32 (onnxruntime CPU) voice=af_heart; Cloud x86 VM, 8 vCPU Intel Xeon, 16 GB RAM, no GPU
benchmark pendingbenchmark pending
Pi 5CPU supported, no device dataUnknownUnknown
CPU onlyMeasured by usUnknownUnknown
Apple SiliconSourced statementEstimate from parameter countEstimate from parameter count
8 GB GPUEstimate from parameter countEstimate from parameter countEstimate from parameter count
12–16 GB GPUEstimate from parameter countEstimate from parameter countEstimate from parameter count
24 GB+ GPUEstimate from parameter countEstimate from parameter countEstimate from parameter count

Each fact is sourced on the model's page. "Unknown" means the source doesn't say. Accuracy: Hugging Face Open TTS Leaderboard.

Key differences (sourced)

Which should you pick?

Check against your hardware in the picker →

New open speech models: will they run on your hardware?

One email a week: new open TTS/STT models, license gotchas, and what we measured on CPUs, Macs, Pis and consumer GPUs.