Spoofed speech from the perspective of a forensic phonetician

While previous work has uncovered the performances of automatic systems when presented with spoofed speech samples, this work looks deeper into these samples from the perspective of an experienced forensic phonetician. From an evaluation of 300 samples, this paper reveals how similar some of the spoofed speech samples are to genuine human speech samples. One speech synthesis method 'sticks out' in this respect by producing speech samples that bear a collection of natural speech characteristics. On the other hand, spoofing methods that have been shown to present problems to automatic systems in past work do not present problems to the forensic phonetician. The main contribution of this paper is a descriptive account of different spoofed speech samples, based on the auditory-acoustic evaluation of a forensic phonetician. This is to spark an awareness of the current possibilities of spoofing methods that perhaps have so far been 'off the radar' for many forensic phoneticians. Such an account can bring great value to the forensic phonetics community as spoofed speech samples could plausibly find their way into forensic casework scenarios.