Case Study: Primetalk achieves scalable, robust Speech Portal for high-volume telephony with Akka

Primetalk uses Typesafe to power Speech Portal

Primetalk builds the Speech Portal, a platform for flexible, user‑initiative spoken dialog systems, but their first Java‑based release proved fragile and didn’t scale. The system struggled with complex NLP workloads, event‑driven real‑time dialog handling, parallel speech recognition, multiple audio sources and concurrent conversations; Java’s thread‑based concurrency and Guice integration led to memory leaks, error‑prone shared state, limited scalability and hard‑to-diagnose configuration problems.

Primetalk migrated to the Typesafe Platform—using Akka and Scala—and developed SynapseGrid to compose functions and actors. The new architecture separates state from threads, favors immutable data and functional components, and eliminates many runtime configuration issues; as a result Speech Portal now supports elastic multichannel scaling, parallel ASR, real‑time media mixing, simpler debugging and much greater resilience.

Primetalk