Managing an ASR Project - From Data to Delivery

Explore the complete lifecycle of Automated Speech Recognition (ASR) project management in this comprehensive lecture delivered by Igor Szöke and Karel Veselý from Brno University of Technology at the JSALT Summer School. Delve into critical deployment challenges spanning from initial data collection through final delivery, including privacy considerations and transcription processes. Examine model pretraining strategies and analyze the advantages and disadvantages of using pre-trained models versus building from-scratch approaches, particularly for low-resource languages. Learn about runtime application considerations, streaming ASR implementation, and deployment strategies that directly impact system performance, latency, and scalability. Gain insights from experienced researchers who have contributed significantly to the KALDI toolkit and led major European Commission and DARPA projects in speech recognition technology. Understand practical decision-making frameworks that influence real-world ASR system development and deployment across various applications and industries.