Lightbend > Case Studies > Primetalk Uses Typesafe to Power Speech Portal

Primetalk Uses Typesafe to Power Speech Portal

Customer Company Size

Startup

Country

Worldwide

Product

Speech Portal
Typesafe Platform
Akka Concurrency Framework
SynapseGrid

Tech Stack

Scala
Akka
Java

Implementation Scale

Pilot projects

Impact Metrics

Digital Expertise
Innovation Output
Productivity Improvements

Technology Category

Analytics & Modeling - Natural Language Processing (NLP)
Platform as a Service (PaaS) - Application Development Platforms
Application Infrastructure & Middleware - Event-Driven Application

Applicable Industries

Software
Telecommunications

Applicable Functions

Product Research & Development
Business Operation

Use Cases

Speech Recognition
Remote Collaboration
Remote Control

Services

Software Design & Engineering Services
System Integration

About The Customer

Primetalk is a new venture in the natural language processing field, focusing on developing spoken dialog systems that offer flexible, interactive, user-initiative dialogs as a human-computer interface for diverse information systems. The company aims to replace the awkward interactive voice response systems (IVR) with user-friendly spoken dialog systems (SDS). Despite being a new player in the field, Primetalk boasts a highly qualified team with substantial experience in software development and scientific research. The team has two years of experience with Scala, over ten years of experience in Enterprise Java development and project management, and five years of scientific research in spoken dialog systems. Primetalk's Speech Portal is a platform for building spoken dialog systems, including components for speech processing and an artificial intelligence layer that enables the dialog system to think and learn. The company's mission is to create a new stack that allows the creation of flexible user-initiative dialogs, addressing the limitations of existing VoiceXML applications that lead to rigid system-driven dialogs.

The Challenge

The goal of Primetalk is to create a completely new stack that allows the creation of flexible user-initiative dialogs. However, this goal is challenging due to several factors. Natural Language Processing (NLP) is a complex field requiring interdisciplinary knowledge in artificial intelligence, neurophysiology, aural physiology, signal processing, computational linguistics, probability theory, and mathematical statistics. Additionally, automatic, large vocabulary continuous speech recognition is a heavy load task that benefits from parallelism, but implementing parallelism is non-trivial. The dynamic event-driven nature of real-time dialog handling presents another challenge, as many events need immediate handling by the dialog logic. These events can include recognition results, audible user interventions, timers, telephony, and system playback. Furthermore, simultaneous playback of different audible media and parallel conversations with multiple participants add to the complexity. Scalability is also a concern, as real-world demand requires handling many channels by a single application instance with zero administration. The initial version of the Primetalk Speech Portal, developed with Java concurrency, faced several drawbacks, including memory leaks, error-prone communication between threads, limited scalability, and unexpected runtime errors due to incomplete configurations.

The Solution

Primetalk sought a better solution and found it in Typesafe’s Akka Concurrency Framework. Akka is a toolkit and runtime for building highly concurrent, distributed, and fault-tolerant event-driven applications on the JVM. Primetalk implemented a pilot project with Akka in Java and found it promising. They also evaluated the Scala programming language during the Akka pilot and were impressed with its capabilities. Scala is a general-purpose programming language designed to express common programming patterns concisely, elegantly, and type-safely. Primetalk then implemented another pilot with Akka in Scala, and the results were so favorable that they decided to base their system on it. Scala and Akka provided Primetalk with the opportunity to rapidly create robust and scalable applications. The separation between state and thread pool in Akka eliminates memory leaks, and immutable data processing excludes locking completely. Functional programming encourages the creation of reusable components, and scalability is perfect as adding more channels does not affect the thread-pool size. The Cake Pattern in Scala allows early detection of configuration problems, reducing runtime errors. As part of their work, Primetalk open-sourced a framework named “SynapseGrid,” a function integration framework that connects ordinary Scala functions into a system by declaring the directed graph of interconnections. SynapseGrid allows flexible function composition, strictly typed message handling in Akka actors, and easy implementation of multiple input/multiple output functions.

Operational Impact

The transition to the Typesafe Platform has provided Primetalk with a strong foundation for building a new generation, scalable, and robust dialog management platform.
The newest release of Speech Portal has many desirable properties, including elastic-scale on multichannel requirements, flexible integration of natural speech processing components, and parallelism of automatic speech recognition.
The dialog manager supports dynamic real-time dialogs, and the real-time mixer can handle many media sources switched in real time.
Stateless functional processing leads to almost zero debugging, and the system is resilient to exceptions both in the Speech Portal and in custom dialog logic.
The use of Scala and Akka has enabled Primetalk to create robust and scalable applications rapidly, with perfect scalability and early detection of configuration problems.