Improving synthetic speech quality using binaural reverberation

The degrading characteristics of synthetic speech, such as minimum phase effects, pitch and voicing errors, and spectral distortions are more evident when the speech is listened to on headphones than when heard in a room over a loudspeaker. Listening to monaural sound in a room over a loudspeaker differs from headphone listening in two major respects: one, a different sound source is presented to each ear (binaural reproduction); and two, the sound source is altered by the room´s acoustics, (reverberation). This paper describes a technique for including the effects of binaural reverberation on synthetic speech heard on headphones. To achieve this effect, the impulse response of a 20´ × 20´ classroom was first measured by applying an electrical pulse to a loudspeaker and recording the resulting room-loudspeaker impulse response as measured by two microphones spaced the ears´ distance apart. These impulse responses were then convolved with the speech and played through each headset channel. Results demonstrate that this process not only suppresses the characteristic distortions of the synthetic speech, but also externalizes the sound source giving the effect of non-headphone listening.