使用电视采集卡进行电视音频处理
我正在寻找一个开源库或框架来处理来自电视采集卡的音频信号。这个想法是检测电视广告位并记录它们发生的时间和频道。我从来没有从事过这样的事情，所以欢迎任何信息、链接、想法。
提前致谢！
编辑：我不关心语言或操作系统。 我想澄清一下，我的目标是检测我工作的公司的具体广告。因此，我认为我应该使用我们广告的指纹并根据捕获的数据对其进行检查。如果您有这方面的信息，也欢迎。
I'm looking for an open source library or framework to process audio signal from a TV capture card. The idea is to detect TV ad spots and register the time and the channel where them happends. I never worked in something like this, so, any information, link, idea is welcome.
Thanks in advance!
EDIT: I don't care about the language or the OS.
I want to clarify that my objective is to detect the specifics ads of the company where I work. So, I think I should use the fingerprints of our ads and check them against the captured data. If you have information about this, is welcome too.
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（2）
最初可以使用虚拟配音之类的东西来捕获视频。
检测广告往往是多种因素的结合：
广告长度往往是 5 秒的倍数，因此可以根据此验证可疑的起点和终点（但在不同的国家/地区可能有所不同）。
声音通常在过渡点完全下降。
空白帧或显着变化也会同时发生。
根据电台的不同，广告开始时可能会有其他徽标和您可以检测到的内容。
显然，并非所有这些事情都是有保证的，您往往最终会根据所有这些因素做出判断，并在稍后进行某种目视检查。
编辑：
要查找特定广告，您需要在广告上存储一些相关信息（如您所说的指纹），例如：
广告长度。
声音级别，可能在多个时间点进行测量。
颜色信息，也许是 RGB 级别，因为它们在整个广告中有所不同。
综合这些因素，您可以想出匹配的“候选者”，但准确率绝不是 100%。
Its possible to use something like virtual dub to capture the video initially.
Detecting ads tends to be a combination of several things:
Ad lengths tend to be multiples of 5 seconds, so the suspected start and end points can be verified against this (can vary in different countries though).
Sound generally drops completely at the transition point.
A blank frame or a significant change occurs also at the same time.
Depending on station there could be other logos and things you can detect when ads start.
Obviously not all these things are guaranteed, and you tend to end up making a judgement based on all these factors, and have some kind of visual check at a later point.
Edit:
To find look for particular ads, you will need to store some pertinant information on the ad (as you say a fingerprint) such as:
Ad length.
Sounds levels, maybe taken at several points in time.
Color info, maybe rgb levels to start with, again as they vary throughout the ad.
As a combination of these things you can come up with 'candidates' for matches though by no means with 100% accuracy.
@Jonathan Barbero：你可以尝试 Gstreamer。
例如：gst-launch -v videotestsrc ！ ffmpeg颜色空间！图像接收器
@Jonathan Barbero: You can try Gstreamer.
ex: gst-launch -v videotestsrc ! ffmpegcolorspace ! ximagesink