Chatterbox Turbo
MIT-licensed 350M-param open TTS — one-step mel decoder, paralinguistic tags, PerTh watermarked.
About
Chatterbox Turbo is Resemble AI's efficiency-focused open-source text-to-speech model, released December 15, 2025. It distills the 0.5B-parameter Chatterbox family down to a 350M-parameter architecture whose speech-token-to-mel decoder was distilled from 10 diffusion steps to a single step, delivering ~6x real-time synthesis on GPU with sub-150ms time to first sound. Paralinguistic tags ([laugh], [sigh], [gasp], [cough], [yawn]) are native to the model, and it does zero-shot voice cloning from about 5 seconds of reference audio with an emotion-exaggeration control.
Everything is MIT-licensed — code and weights — via GitHub (resemble-ai/chatterbox) and Hugging Face (ResembleAI/chatterbox-turbo), installable with pip install chatterbox-tts, and deployable on Replicate, Fal, Modal, and RunPod. Every output carries Resemble's imperceptible PerTh neural watermark for provenance. Independent signal: ~1,020 Elo on the Speech Arena; Resemble's published blind evals show Turbo beating ElevenLabs Turbo v2.5 in 65.3% of comparisons and tying Cartesia Sonic 3 at 49.8% — with the losses published too, which is more credibility than most vendor tables earn.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Independent: Speech Arena Elo ~1,020 (provider-voice).
Provider-reported blind evals: 65.3% win rate vs ElevenLabs Turbo v2.5, 49.8% tie vs Cartesia Sonic 3.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Verified directly against the public GitHub repo (resemble-ai/chatterbox) and Hugging Face weights (ResembleAI/chatterbox-turbo); MIT license on code and weights is checkable. Independent Speech Arena Elo corroborates quality. Docked for: (1) every output carries an indelible PerTh watermark — a provenance feature for some, a constraint for others; (2) blind-eval figures are provider-run, though unusually transparent.