Experiments on a Channel Error Protection Scheme for Distributed Speech Recognition

Low recognition rates are often experienced in speech recognition over wireless networks where propagation environments may cause high transmission error rates. One way to reduce the sensitivity towards transmission errors is to use a distributed speech recognition architecture (DSR) which eliminates the speech channel and instead uses an error protected data channel to transmit a parameterised representation suitable for speech recognition of the speech. Within the ETSI-DSR standard, two quantised melcepstral frames – each of 10 ms duration are grouped together and protected with a four-bit Cyclic Redundancy Checking (CRC) forming a frame-pair. However, this scheme increases the Frame Error Rate (FER) over an error-prone transmission channel. To overcome this, the paper presents a one-frame architecture in which a four-bit CRC is calculated to protect each frame independently. This scheme results in a lower overall probability of one frame in error at the cost of only a marginal increase in data rate from 4800 bits/s to 5000 bits/s. A number of recognition experiments have been conducted on two different recognition tasks, digits and city names, to verify the introduction of the one-frame CRC protection scheme for a number of simulated transmission channel bit-error rates (BER) ranging from 0 (no transmission channel involved) to 2 %. Experimental results verify that the one-frame error protection scheme is more robust against channel errors.