We have a large number of audio files we need to make transcripts from. Are there any affordable, automatic-transcription software tools available or are we stuck using human transcriptionists?

Getting accurate transcripts from audio files through software can be tough, but it is getting easier. There are shareware and commercial WAV-to-text programs available for converting audio files to text. Two examples are here and here. The Speech at CMU project page provides information about open source speech-recognition resources. And there are large commercial systems aimed at providing audio-mining capabilities as part of enterprise search systems available from places like DocSoft. Microsoft's Speech API (SAPI) and software development kit (SDK) can be used to build speech-recognition applications in .Net. Between the large-scale proprietary tools and developing your own speech-recognition software to transcribe audio files lies the middle ground of improving the performance and productivity of human transcriptionists. One of the most easily implemented and effective ways to speed up transcription work is to make use of automated-typing software like Dragon Naturally Speaking or ViaVoice in a workflow where the operator listens to the audio file and repeats the text into the voice-recognition software. We've found that this is often faster than listening and typing by hand. Depending on the clarity of the recording and the noise in the environment, it is sometimes possible to play the recording into the microphone and be successful in creating an accurate transcript. Your mileage will vary depending on the playback speakers and microphone available.