<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Ted Malone on Medium]]></title>
        <description><![CDATA[Stories by Ted Malone on Medium]]></description>
        <link>https://medium.com/@ted-malone?source=rss-83b38ad19533------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*fqNgFlZpHlx0B12XnlEIyA@2x.jpeg</url>
            <title>Stories by Ted Malone on Medium</title>
            <link>https://medium.com/@ted-malone?source=rss-83b38ad19533------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 11:19:25 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ted-malone/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Identifying TV commercials with audio + Live web demo]]></title>
            <link>https://ted-malone.medium.com/identifying-tv-commercials-with-audio-live-web-demo-b310d1d421ee?source=rss-83b38ad19533------2</link>
            <guid isPermaLink="false">https://medium.com/p/b310d1d421ee</guid>
            <category><![CDATA[dvr]]></category>
            <category><![CDATA[tv-shows]]></category>
            <category><![CDATA[tv-commercials]]></category>
            <category><![CDATA[audio]]></category>
            <category><![CDATA[tivo]]></category>
            <dc:creator><![CDATA[Ted Malone]]></dc:creator>
            <pubDate>Wed, 18 Nov 2020 17:51:45 GMT</pubDate>
            <atom:updated>2020-11-18T17:51:45.724Z</atom:updated>
            <content:encoded><![CDATA[<p>For the past few months I have been working closely with a startup in the audio processing space called <a href="https://cyberresonance.com/">Cyber Resonance</a>. CyRes has two very cool, cloud-based processing services. Today, I’m writing about my experience with one of them — <strong>Media Detective</strong> (mD). With mD, you can quickly and easily identify TV ad breaks using a combination of audio processing and artificial intelligence.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/695/1*asYRrrsdUbkO_dC1R5nlig.png" /></figure><p>In the audio waveform above, you can see several audio drops through an ad pod (a series of individual ads). You can also see an audio drop earlier in the audio stream which was actually part of the program content, not an ad break.</p><p>From a market standpoint, there are a variety of things you can do once you’ve identified the TV commercials in a video stream. At CyRes, we’ve identified several key use cases:</p><ul><li><strong>Commercial Removal</strong> — Allow a consumer to watch the programming content and skip over the ads</li><li><strong>Commercial Replacement</strong> — Identify the ad breaks and refresh them with new or targeted content</li><li><strong>Commercial Identification</strong> — After identifying the individual ads, use fingerprinting in order to track and report which ads were shown</li><li><strong>Storage Optimization </strong>— For video libraries that are created by capturing live TV feeds, identifying and removing the commercial can dramatically reduce storage space/cost.</li></ul><p>I’m not a programmer by day, but I do still write code as a hobby. So, I decided to build a web demo so potential customers could try out Media Detective to see how it works, it’s output format, etc. I’m planning to write a few small articles to describe the technologies I used to build the demo, and I will hopefully get around to publishing the code on GitHub once I clean it up a bit.</p><p>You can access the demo website here: <a href="https://mddemo.cyberresonance.com/"><strong>mD Demo</strong></a> The demo includes a few sample audio tracks from some popular TV programs, and it also allows you to upload and test your own! The demo has been tested to work with .wav, .m4a or .flac audio files. In order to test it, I used <a href="https://sourceforge.net/projects/kmttg/">KMTTG</a> to transfer some recordings from a TiVo DVR, and then stripped the audio into it’s own file with <a href="https://ffmpeg.org/">ffmpeg</a> using this command:</p><p><strong><em>ffmpeg -hide_banner -i INPUT_FILE.ts -vn -sn -ac 2 -y Output.m4a</em></strong></p><p>Once the audio track is ready, you can send it to mD for processing. A 30-minute program takes about 2 minutes to process, and a 1-hour program takes about 4 minutes. The results come back in a nice looking table like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/834/0*1w58ysXTI4Kqr_cX" /></figure><p>After processing, there is also a link that let’s you see the raw JSON that is returned from Media Detective. If you’re the kind of person who likes looking at JSON, here you can see the raw data that went into the table above — <a href="https://mddemo.cyberresonance.com/fetch-md?taskID=sqTg6aMzoslUVKP8bh4MWpT0philtinU">RAW DATA</a></p><p>If you end up trying mD with your own sound track, please leave a comment and let me know how the results looked to you. The default settings don’t always work perfectly. In production, we use additional metadata (show name, genre, etc.) to increase accuracy. We’ve also found that it helps if you can include a small amount of pre and post-roll to help the algorithm identify commercial breaks at the beginning and end of a program.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b310d1d421ee" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>