Sped_up_audios_wtimestamps Now
: A 2025 paper that introduces a data-driven approach using the Canary model. It uses a <|timestamp|> token to predict start and end times for words with high precision (80–90%), even as audio characteristics change.
: This paper explores the effectiveness of combining transcripts with pitch-normalized, time-compressed speech. It specifically looks at how speed impacts user comprehension and the accuracy of machine-generated text alignments.
: This 2024 paper improves timestamp precision for OpenAI's Whisper model. It addresses "unsharp" timestamps caused by pauses or rapid speech by adjusting the model's tokenizer and using cross-attention scores for alignment.
WhisperX: Automatic Speech Recognition with Word ... - GitHub
: A 2025 paper that introduces a data-driven approach using the Canary model. It uses a <|timestamp|> token to predict start and end times for words with high precision (80–90%), even as audio characteristics change.
: This paper explores the effectiveness of combining transcripts with pitch-normalized, time-compressed speech. It specifically looks at how speed impacts user comprehension and the accuracy of machine-generated text alignments.
: This 2024 paper improves timestamp precision for OpenAI's Whisper model. It addresses "unsharp" timestamps caused by pauses or rapid speech by adjusting the model's tokenizer and using cross-attention scores for alignment.
WhisperX: Automatic Speech Recognition with Word ... - GitHub