July 2020 Summaries

ASR evaluations are created to help developers troubleshoot speech recognition issues and improve performance, identifying commonly misrecognized words for better customer experience. A utility has been developed, open-sourced on GitHub, to simplify the evaluation process by performing pre-processing normalization of text and going beyond Word Error Rate (WER). This tool can calculate various metrics, including WER, Levenshtein Distance, and text comparison to visualize differences. It provides an easy-to-use command-line interface for running evaluations with original and generated files. The utility aims to make ASR evaluations more convenient and time-efficient for developers, offering free trial credits for exploring conversational intelligence solutions.

Congratulations on considering Conversational Intelligence as part of your product offerings, this technology can unlock significant value by analyzing employee and customer conversations in real-time, providing insights to improve productivity and engagement. To determine the best solution for your business, consider whether you need real-time transcription or closed captioning capabilities, how you plan to monetize existing data, whether you want a finished product or flexibility to add capabilities to your own offering, what type of content you need to analyze, and how cost will impact your decision as you scale. Additionally, weigh the pros and cons of building versus buying a solution, consider the importance of training data, and think about the on-going conversational intelligence roadmap and its alignment with your other product offerings or AI capabilities across the company. Ultimately, prioritize the user experience to ensure a seamless integration into existing workflows and maximize engagement and value generation.

Choosing the right integration approach and providing high-quality speech data is crucial for accurate results when using Voice or Audio APIs. The guidelines cover key considerations such as choosing the right API, audio chunk size, background noise, multiple people in a single channel, calibration, telephony API best practices including SIP over PSTN, audio codecs, transmission protocol, secure SIP, and Real-time WebSocket API best practices. These recommendations aim to optimize accuracy and efficiency while considering latency, network reliability, and robustness.