CleanSpeak vs Detoxify

CleanSpeak is a configurable moderation toolset for services that need to filter or review user content. It filters text, images, and video, and can moderate audio but does not filter it. Moderation queues let teams review and escalate material, while custom rules can automate actions by content source, rate users, and trigger actions from those ratings. Editable blocklists detect negative terms and phrases, including altered forms intended to bypass filters. The product includes 22+ base language lists that customers can change or supplement. Tunable text, image, and video models integrate Google Perspective and SightEngine; reports cover moderator activity, content searches, and term or model frequencies. CleanSpeak is available as managed cloud service or as server software installed and maintained by customers. Its synchronous REST APIs return JSON for real-time integration across content filtering, moderation, and user management. The company says one server can filter 10,000 or more messages per second. It serves use cases in apps and gaming, retail and hospitality, and enterprise settings. The advertised trial lasts 30 days and requires no credit card; pricing is available on request.

Detoxify is a free, self-hosted library that provides trained models and code to classify toxic comments. Its original, unbiased, and multilingual models are based on BERT, RoBERTa, and XLM-RoBERTa, respectively. Depending on the model, outputs include categories such as threats, obscenity, insults, identity attacks, and explicit sexual content. The multilingual model is intended for English, French, Spanish, Italian, Portuguese, Turkish, and Russian. The project describes research, fine-tuning on carefully constructed representative datasets, and helping moderators flag harmful content faster as intended uses. It can be installed with pip and called from Python to predict labels for one string or a list. Inference uses CPU by default and accepts a PyTorch device such as CUDA. The project warns that profanity-associated words may be labeled toxic regardless of tone or intent, potentially biasing results against vulnerable minority groups. Inference uses Transformers and PyTorch Lightning; training also requires the Kaggle API to download data. API access can be requested by email. The library is distributed under Apache-2.0.