Sharper summaries on every message
Summaries got a quality upgrade and are more reliable when audio is difficult. Word count and processing time moved out of the email and into the dashboard.
Summarization moved to a newer model, and it is now given the transcription's own detected language, speaker count, confidence, and duration, so it has more to work with when the audio is difficult. Transcription and file transfers retry with a widening delay on transient failures, so a momentary blip no longer costs a voicegram. The recorder captures with noise suppression, echo cancellation, and automatic gain in a single channel, which makes confidence scores more consistent from one browser to the next. Notification emails gained an intent badge and paragraph breaks for single-speaker transcripts.
