Case Study: Speak cuts signal-production time and boosts model accuracy with Labelbox

Speak cuts signal-production time by nearly 50% with Labelbox

Speak, a fast-growing AI language learning app, faced the challenge of scaling its high-quality data operations to train and evaluate its speech recognition and language models. As its user base grew, managing diverse audio data and producing accurate, nuanced signal for accents and dialects became a major hurdle. They turned to the vendor Labelbox to address this.

By implementing the Labelbox platform, Speak centralized its data management, quality control, and collaborative feedback. This solution created an automatic labeling loop that daily evaluates and retrains its models on new production data. As a result, Speak cut its signal-production time by nearly 50%, accelerated model development, and achieved model accuracy improvements of up to 35%, allowing for faster release of new features and languages.

Speak