March 2024 Summaries

Speechmatics introduces a bilingual Spanish and English transcription model that can seamlessly switch between languages in real-time audio streams or batch files. This innovative solution addresses the challenges of accurately transcribing varying accents or dialects within multilingual environments, particularly crucial for businesses navigating diverse linguistic regions. The new model enhances accessibility and inclusivity by allowing users to understand and transcribe over 2 billion people in a single API call.

Speechmatics has recently improved its speech recognition technology for Norwegian by reducing transcription errors by 40%. The company tackled challenges such as acoustic variations, diverse vocabulary, and grammatical intricacies to achieve this improvement. Key factors included addressing regional dialects across five regions in Norway and the dual written forms of Bokmål and Nynorsk. Speechmatics used three main levers for improving accuracy: model improvements, better data, and language-specific enhancements. The result is a more accurate Norwegian speech recognition system that benefits downstream tasks like summarization and translation.