Case Study: Largest VOIP Provider achieves near human-level speech transcription accuracy with Quantiphi

Largest VOIP Provider - Customer Case Study

Quantiphi worked with the Largest VOIP Provider, one of the largest VOIP providers for financial markets with operations in 34 countries. The customer needed a speech recognition solution that could handle financial jargon, overlapping speakers, short trade “shouts,” multiple accents, and millisecond response times, while also dealing with a lack of labeled training data.

Quantiphi built a domain-adapted speech recognition solution using deep learning, including stacked CNN and RNN acoustic models in TensorFlow, human-in-the-loop data labeling, distributed training on Google CloudML with GPU acceleration, and an RNN-based re-scoring language model. The result was a state-of-the-art transcription system with near human-level accuracy, enabling compliance, audit, surveillance, and trade automation use cases.