PodZinger is a multimedia search engine that uses speech recognition to increase the relevance of the search results. While other search engines use only the metadata (title, description, tags) to index audios and videos, PodZinger treats each file like a text document. It shows text snippets relevant to your queries and you can click to a word from the snippet to go directly to that part of the file.
Their speech recognition technology works very well (only for English and Spanish), but the index is small, as it contains only podcasts. This technology would be a good addition to sites like Google Video or YouTube, because the metadata can be insufficient or misleading. Google Video has the option of adding captions, but very few people will use it.