Open weightsResemble AI/10 trustVerified

Chatterbox Nano

Name: Chatterbox Nano
Rating: 8

Resemble AI's 110M-parameter edge TTS: Turbo's single-step architecture shrunk to run 3x real-time on an 8-core CPU — zero-shot cloning and [laugh]-style tags included.

Visit official page →

About

Chatterbox Nano is the smallest member of Resemble AI's open Chatterbox TTS family, built for the deployments the 350M Turbo model can't fit into: edge devices, mobile apps, on-prem servers, and air-gapped boxes. It keeps Turbo's headline trick — a distilled speech-token-to-mel decoder that collapses the old 10-step process into a single step — but on a streamlined 110M-parameter backbone. Resemble AI reports 10x real-time on GPU and 3x real-time on an 8-core CPU, which puts usable voice agents on hardware with no accelerator at all. It is trained on the same dataset as Turbo, so the expressive range carries over: zero-shot voice cloning from 10+ seconds of reference audio and native paralinguistic tags ([laugh], [sigh], [cough], [chuckle], [gasp]) rendered in the cloned voice. Every output carries Resemble's imperceptible PerTh watermark. Code and weights are MIT licensed with no commercial restrictions, installable via pip (chatterbox-tts), with weights on Hugging Face.

Specs

Context window

Not applicable — text in (+ optional 10s+ reference clip for cloning), 24 kHz mono speech out; English

Pricing

Free / open release

Free quota

Fully open — MIT-licensed weights, self-host via pip (CPU or GPU)

API access

No hosted model API — self-host via pip (chatterbox-tts); Resemble AI also sells a managed TTS service (sub-200ms latency claimed)

Weights

Open weights

Released

Not disclosed

Benchmarks

Provider-reported: 3x real-time on 8 CPU cores; 10x real-time on GPU
No independent benchmarks found

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

voice-agentsedge-aitext-to-speechvoice-cloningivr

Trust score — 8/10

/10 trustVerified

Scored 8/10 (TIER verified, metrics-v1). Evidence: Resemble AI's open Chatterbox family (27k GitHub stars); public model card with architecture details; pip-installable with HF weights. Strengths: MIT license with zero commercial restrictions, genuinely CPU-viable TTS with Turbo's tag vocabulary and cloning carried over, imperceptible PerTh watermarking. Docked for: (1) speed and quality claims are provider-reported with no independent benchmarks found; (2) English-only and the managed-service upsell means the open model may lag future paid improvements. License: MIT.

How the trust score is built

Full methodology →

Adoption · stars/downloads, log-scaled vs peers: 8.0/10
Maintenance · recency, releases, issue load: 8.0/10
Openness · license + docs: 10.0/10
Manual review · sanity read + red-flag check: 8.0/10

· 110M params, English, MIT; single-step distilled decoder from Turbo lineage
· Provider-reported 10x real-time GPU / 3x real-time on 8 CPU cores; zero-shot cloning from 10s+ reference audio
· Native paralinguistic tags ([laugh], [sigh], [cough], [chuckle], [gasp]); every output carries imperceptible PerTh watermark
· Chatterbox family repo at 27k stars; pip-installable (chatterbox-tts), weights on Hugging Face