Quantifying the Value of Subjective and Objective Speech Intelligibility Assessment in Forensic Applications

摘要

Transcription from lawful interception is an important branch of forensic phonetics. Signals in that application context are often degraded, thus the transcript may not reflect what was really pronounced. In order to decide whether a given transcript generated from a lawful interception exercise reflects the views of the speakers instead of the transcriber's, an objective speech intelligibility measurement method is required. Usually, the intercepted signal can be affected by both speech intrinsic distortion and background/environmental noise distortion. Unfortunately, the original clean speech is never accessible to the forensic expert, who therefore must draw his assessment from the only available, distorted, signal. Consequently, the only way to assess the level of accuracy that can be obtained in the transcription of poor recordings is to develop an objective methodology for intelligibility measurements. This paper addresses the issue by using three different objective approaches - namely the Signal-to-Noise ratio weighted with the "A" curves (S/NA), the Articulation Index (AI) and the Speech Transmission Index (STI) - to evaluate the intelligibility of a given signal. All of the three approaches were exercised with different types of noise, yielding results to be compared with speech intelligibility scores from subjective tests. The outcome gives high correlation evidence between objective measurements and subjective evaluations. Therefore, the proposed methodology is deemed rather useful to establish whether a given intercepted signal can be transcribed with sufficient reliability.