- claude
- yapay-zeka
- video-duzenleme
- ses-analizi
- acik-kaynak
- earshot
- claude-code
Claude Can't Hear Your Video — Earshot Fixes That
Earshot analyzes a video's audio and produces a report Claude can actually read. Here's how this free, open-source tool works.
Watch on Instagram
Editing video with Claude is increasingly popular, but there's a significant blind spot in that workflow: Claude can't hear audio.
The Problem: Claude Is Deaf
You can hand Claude a video and ask it to add effects — keyboard sounds, ambience, music — but Claude has no way of knowing whether those effects actually fit the existing audio or your speaking tone. It places the effect without being able to answer the question "does this work?" That gap directly affects the perceived quality of the final video.
What Is Earshot?
Earshot is a free, open-source tool published on GitHub. Its core claim: "I'm here to be Claude's ears." The tool analyzes a video's audio layer and converts it into a structured report that Claude can read. The report includes:
- Average loudness (LUFS): The overall volume level of the video
- Peak points: Sudden spikes in volume
- Frequency distribution: Balance between bass, mid, and high frequencies
- Mono/stereo status
- Effect detection: Audio effects already present in the video
Beyond the raw data, Earshot benchmarks the analyzed video against 7 professional promo films and produces a scorecard, giving you a numerical sense of how close your video is to industry standards.
A Real Measurement: −23.9 LUFS
I ran the tool on one of my own videos. The result: an average of −23.9 LUFS. The typical target is around −14 LUFS, meaning my videos were roughly 10 dB quieter than they should be. That gap might sound minor, but for viewers it creates a real problem: they have to keep adjusting their phone volume, which breaks focus.
The report also revealed peak fluctuations reaching −4.4 LUFS at certain moments — sudden spikes that make for an uncomfortable listening experience in uncompressed videos.
Feeding the Report to Claude
I pasted Earshot's technical report directly into Claude Code and used this prompt: "Interpret this technical analysis in plain language I can understand." Claude summarized the output under these headings:
- Audio is too quiet
- Too many emphasis spikes
- Audio is entirely mono
- Bass frequencies are nearly absent
- Mid frequencies are a bit harsh
Below the summary, Claude offered concrete suggestions for fixing each issue — a useful and accessible output even for someone without audio engineering experience.
Limitations
The tool isn't perfect. It misidentified some of my vocal emphasis as sound effects. When I raised my voice dramatically while narrating, Earshot tried to interpret it as an effect. These false positives are more likely to appear in content where the narrator's delivery is highly dynamic.
Who Should Use It?
If your microphone causes problems, your sound effects never quite sit right in the mix, or you simply can't nail your video's overall audio level, Earshot is a solid free starting point. It's open-source and integrates directly into a Claude-based editing workflow.
To install it, follow the README on the GitHub page. You can also paste the repository link directly into Claude and tell it to install the repo — that works too. For ready-made prompts, follow the instructions in the video description.