MacWorld Review of the Soundbooth Beta and Speech Search
Here’s a link to the MacWorld First Looks at the beta of the next version of Soundbooth.
All in all it’s a good review. The one thing I did want to respond to were the reviewer’s comments about the new speech search feature.
The reviewer’s results in creating speech to text transcripts were only about 50% accurate. The accuracy of the transcripts generated by the speech engine will vary depending on the quality of the audio and the type of speech. It will work best with broadcast quality audio like news feeds. One way to test this is to pick up a newspaper or magazine and record your own voice reading a news article. There you will see results in the 90% accuracy range. (Any USB mic will work fine for this and if you don’t have one Blue Microphones makes a couple very nice affordable USB mics called the Snowball and the Snowflake).
Now as the reviewer noted even with a 50% accurate transcript the speech engine will capture enough keywords to enable you to easily search through your audio stream. This really speaks to the heart of what makes this new feature so valuable. Speech search is about speeding up the editing process by enabling you to easily locate specific words in your audio clips. Also, with the new metadata panel you can easily correct words in the transcript. For each word there will be a confidence rating displayed giving you an indication of whether the word is accurate or not.
The other thing to note about the speech engine is that is has different language models including ones to cover different dialects. This will give you more accurate results when you have speakers with an Australian or British accent. For the beta we only ship the American English language model, but other language models will be available with the shipping release.