How to Fix Choppy Audio and Video Cuts in Riverside
When Riverside removes gaps and silences from your recording, the cuts sometimes sound choppy — words get clipped, or the edit feels abrupt. The audio and video themselves are fine. The problem is in the transcription alignment.
Riverside transcribes your recording at the word level, assigning timestamps to the start and end of each word. It then uses these word boundaries to detect gaps and silences. When the transcription places a word boundary slightly before or after the actual audio, the resulting cut lands in the wrong spot — clipping the beginning or end of a word.
Why This Happens
Automatic speech recognition isn't perfect. It identifies words and maps them to timestamps, but those timestamps can be off by fractions of a second. When Riverside's gap removal relies on those timestamps to decide where silence begins and ends, a misaligned word boundary means the cut happens mid-word instead of in actual silence.
This isn't an audio/video sync issue. The tracks are perfectly in sync. The issue is that the transcription's word-level timestamps don't line up precisely with the audio waveform.
How to Fix It
- Switch to the timeline view in the Riverside editor
- Zoom in on the cut that sounds choppy — you'll see a shaded area where the gap was removed
- Adjust the edges of the cut — drag the boundary of the shaded area to shorten the removed section, giving the word more room to play fully
- Preview the edit to confirm the word is no longer clipped
The key is to look at the waveform. If you can see that the cut is eating into the audio of a word (the waveform is still active where the cut begins), pull the cut boundary back until the word plays cleanly.
Tips
- Zoom in as far as you can — these adjustments are often just a few frames
- Focus on the waveform, not the transcript text — the waveform shows you exactly where audio starts and stops
- If many cuts are choppy, consider reducing the aggressiveness of gap removal or switching to manual editing for that section
