Realistic listening improved by adding fine structure

Speech communication often takes place in noisy and reverberant environments. Unfortunately, cochlear implant (CI) users perform significantly poorer than normal hearing listeners under adverse conditions. We hypothesize that the noise susceptibility of CI users is largely due to the lack of encoding the fine structure cue. Most CI speech processing algorithms extract and encode the slowly varying amplitude modulation in speech (the envelope), while the rapidly varying fine structure is replaced with a fixed rate carrier. We have developed a new speech processing algorithm, called Frequency-Amplitude-Modulation-Encoding (FAME), which codes both amplitude and frequency modulations. Normal hearing listeners were presented with IEEE sentences in the presence of an interfering voice as a function of the number of noise bands (information channels) and signal-to-noise level. Sentence recognition with a competing talker was generally improved with the addition of fine structure, and by as much as 60% with 4 frequency bands (i.e., from 20%–80% at a 20 dB signal-to-noise level). This result suggests that the fine structure is important for speech recognition under realistic situations and should be encoded in future cochlear implants.