We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Hacker News·Jun 12·5 min read
Removing 'um' from a recording is harder than it sounds
TL;DR
Editing out disfluencies like 'um' and 'uh' from audio recordings is a tedious task. To address this, a tool called erm was developed, which automates the removal process.
✦ Why It Matters
Engineers can leverage the erm tool to automate audio editing, improving efficiency and output quality.
Key Takeaways
- Explore using the erm tool for audio editing tasks.
- Understand the limitations of naive disfluency removal methods.
- Consider the impact of disfluencies on audio quality.
- Utilize JSON cut lists for better audio management.
Full Summary
Disfluencies, such as 'um', 'uh', and their elongated forms, are common in spoken English and can detract from audio quality. To simplify the editing process, the tool erm was created, which automates the removal of these filler words.
The naive approach involves transcribing audio with word-level timestamps and cutting out identified disfluencies using ffmpeg, but this method only achieves about 60% accuracy and often results in poorer audio quality. Instead, erm employs a more sophisticated methodology that enhances the audio output.
It generates a cleaned .wav file alongside a JSON cut list that details the removed segments. The improvements in audio quality make it a valuable tool for anyone working with voice recordings.
This tool can save time and effort for audio editors and researchers alike.