This paper presents a novel method to extract Speech Transmission Index (STI) from reverberated speech utterances using an artificial neural network. The convolutions of anechoic speech signals and simulated impulse responses of rooms of various kinds are used to train the artificial neural network. A time to frequency domain transformation algorithm is proposed as the pre-processor. A multi-layered feed forward neural network trained by back-propagation is adopted. Once trained, the neural network can accurately estimate Speech Transmission Index from speech signals received by a microphone in rooms. This approach utilises a naturalistic sound source, speech, and hence has potential to facilitate occupied measurement.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.