Voicebox

Customer Case Study

Voicebox

Voicebox is leader in conversational AI development, giving you everything you need to build secure speech recognition applications.

Vertical Use Case

Automatic Speech Recognition (e.g. voice assistants)

Technical Use Case

• Ingest and ETL
• Deep Learning

The Challenges

  • Accuracy: They need to engage with criteria components (points of interest, addresses, music, etc.). Ex:”Play the latest song from Coldplay” and then it finds the song and plays. This data is very volatile, meaning it is constantly changing. But their ASR and NRU engines need that information and they need to augment/normalize/aliasing to deliver accurate results.
  • Latency: The response time of getting an answer can impact user experience. They want to reduce it as much as possible.
  • Before Databricks: They have been building car applications for over 10 years. New tech over that time. Everything was done manually, increase in complexity, etc. These factors contributed to accuracy and latency issues, not to mention time to market of new products.

The Solution

Voicebox uses Databricks to build, schedule, and run their production data pipelines that feed into their deep learning models. They are able to automate the following production jobs which has saved significant amounts of time and resources:

  • Criteria Pipelines to automatically update their models so that their engines can accurate interpret things.
  • Latency Pipeline to accelerate processing performance of data feeding into their models.

Business Benefits: Databricks has allowed Voicebox to process data much faster and more reliably, allowing them to build and iterate on their core engines much quicker resulting in increased marketing opportunities (beyond the auto market) and revenue.

Databricks has greatly reduced our data engineering complexity so that we can focus on bringing innovative products to market must faster.

Peyvand Khademi, Director, Data Platform and Services at VoiceBox Technologies