“If you have people transcribe conversational speech over the telephone, the error rate is around 4 percent,” says Xuedong Huang, a senior scientist at Microsoft, whose Project Oxford has provided a public API for budding voice recognition entrepreneurs to play with. “If you put all the systems together—IBM and Google and Microsoft and all the best combined—amazingly the error rate will be around 8 percent.” Huang also estimates commercially available systems are probably closer to 12 percent. “This is not as good as humans,” Huang admits, “but it’s the best the speech community can do. It’s about as twice as bad as humans.”

http://www.wired.com/2016/04/long-form-voice-transcription/