Changelog

Enhanced Transcription Features & Speech Processing

1. Gladia Transcription Enhancements: Improve transcription accuracy and performance with new GladiaTranscriber features:

   • region: Choose between us-west and eu-west for optimal latency and data residency compliance
   • receivePartialTranscripts: Enable low-latency streaming transcription for real-time conversation flow
   • Enhanced language detection with support for both single and multiple language modes

2. Advanced Deepgram Controls: Fine-tune speech recognition with enhanced DeepgramTranscriber settings:

   • eotThreshold: End-of-turn detection threshold for precise conversation boundaries (e.g., 0.7)
   • eotTimeoutMs: Maximum wait time for end-of-turn detection in milliseconds (e.g., 5000ms)
   • eagerEotThreshold: Early end-of-turn detection for responsive conversations (e.g., 0.3)

3. AssemblyAI Keyterms Enhancement: Boost recognition accuracy for critical terms with AssemblyAITranscriber.keytermsPrompt:

   • Support for up to 100 keyterms, each up to 50 characters
   • Improved recognition for specific words and phrases
   • Additional cost: $0.04/hour when enabled

4. Speechmatics Custom Vocabulary: Enhance recognition accuracy with SpeechmaticsCustomVocabularyItem:

   • content: The word or phrase to add (e.g., “Speechmatics”)
   • soundsLike: Alternative phonetic representations (e.g., [“speech mattix”]) for better pronunciation handling

5. Word-Level Confidence: Access detailed transcription confidence data with CustomLLMModel.wordLevelConfidenceEnabled, providing word-by-word accuracy metrics for quality assessment and debugging.

6. Enhanced Message Metadata: Store transcription confidence and other metadata in UserMessage.metadata, enabling detailed analysis of transcription quality and user speech patterns.

Transcription Improvements