Langdrift
ActiveDoes the translation carry the same cognitive weight?
An open source tool that measures cognitive divergence between speech translations using brain-predicted neural activation. It compares how similarly the original and translated audio activate the human brain, then returns a drift score. The higher the drift, the greater the cognitive difference.
What it does
Langdrift uses Meta FAIR's TRIBE v2 model, trained on fMRI data from 720 people, to map audio onto roughly 20,000 cortical vertices. It then calculates a drift score via Pearson correlation between the original and translated activation. Run langdrift run original.mp3 translation.mp3 and get a number back.
Key features
- CLI-based comparison of original versus translated audio.
- Threshold presets: hard (10% / 5%), medium (20% / 10%), low (30% / 20%).
- Supports .mp3, .wav, .m4a, .flac, .ogg, .aac and .opus.
- JSON output for automation and CI pipelines.
- One-time setup wizard for Python, Modal auth and HuggingFace credentials.
Stack & status
TypeScript / Node.js CLI on top of a Python backend running on Modal (A10G GPU). Core ML is Meta FAIR TRIBE v2 (Wav2Vec-BERT 2.0 encoder), with Pearson correlation for scoring. Apache 2.0 licensed (model weights CC BY-NC 4.0), in active development.