Word-Level Timestamps
A technique that provides the start and end times of each word in speech recognition output. It is a foundational capability for subtitle generation, keyword search, and audio editing.
Timeline (last 90 days)
Internalizing speaker diarization and word-level timestamps into the ASR model typically requires jointly predicting speaker labels and time boundaries at decoding time, with the benefit of eliminating the error accumulation and extra latency of a cascaded pipeline.
Jointly predicting speaker labels and time boundaries requires additional transcription corpora with speaker and time boundary annotations for training, and joint decoding may slightly affect the accuracy of plain text recognition.
All Facts (2)
Internalizing speaker diarization and word-level timestamps into the ASR model typically requires jointly predicting speaker labels and time boundaries at decoding time, with the benefit of eliminating the error accumulation and extra latency of a cascaded pipeline.
50%UnverifiedJointly predicting speaker labels and time boundaries requires additional transcription corpora with speaker and time boundary annotations for training, and joint decoding may slightly affect the accuracy of plain text recognition.
50%