tts.spqw.net

Open-source text-to-speech — ranked for this CPU box, with listen samples.

Leaderboard · CPU · 5 Oct 2026

8-core CPU, no GPU. Speed is × realtime measured here on the same sentence.

# Model On this box
1 Supertonic 3
99M · ONNX · OpenRAIL-M · 31 languages · 10 voices
 Default · ~5× realtime · 44.1 kHz
2 Kokoro-82M
ONNX INT8 · Apache 2.0 · presets
 ~0.9× realtime · highest-rated small model in blind tests
3 Chatterbox Nano
110M · MIT · clone + tags
 ~0.4× realtime · voice cloning
— Piper
ONNX · MIT · many voices
 Lowest latency, plainer voice
— Pocket TTS · MOSS-TTS-Nano
~100M · CPU streaming · cloning
 Not yet measured here

Leaderboard · GPU

Not runnable fast on this box. Ordered by open-weights arena rating.

# Model Notes
1 Breeze TTS 2
Aug 2026 · non-commercial weights
 Best quality · voice design · 12 GB+ GPU
2 Fish Audio S2 Pro
80+ languages · non-commercial weights
 Best cloning · inline emotion control
3 Step Audio EditX · Voxtral TTS
2026 open weights
 Strong runners-up
— Chatterbox Turbo / Multilingual
MIT
 Best permissive-licence cloning

Supertonic 3

99M · ONNX · CPU · preset voices · 31 languages

F1

English · female

Hello. This is Supertonic 3, the new default voice engine running on this cloud computer.

M1

English · male

It runs several times faster than real time on a plain CPU, with no graphics card at all.

F3

English · female

Weekly reviews keep this page current as new open models are released.

M3

Korean · male

안녕하세요. 이 목소리는 같은 모델로 만든 한국어 예시입니다.

Kokoro

82M · ONNX INT8 · CPU · presets · most natural English presets

af_heart

American · female

Hello — this is Kokoro running on this cloud computer.

am_adam

American · male

State of the art text to speech can still run fast on CPU.

af_sky

American · female

Weekly reviews keep the default stack current.

bm_george

British · male

These demos are served from tts.spqw.net.

Chatterbox Nano

110M · CPU · zero-shot clone from a short reference

nano · af_heart ref

clone

Chatterbox Nano is a small English text to speech model running on CPU for tts.spqw.net. This clip clones a short reference voice.

nano · bm_george ref

clone · [chuckle]

This second Nano sample uses a different reference voice [chuckle], and it still speaks a short informative English line.