P Forens an experi J. Sarwo Department of Engineering Ph PACS: 43.60.DH , 43.70.BK, 43.72.UV ABSTRACT Since 2008, intercepted speech comm first case of using this evidence in the tion Commission (KPK). To take th Speaker Identification Forensic Exam needed. The first group is the interce cording of an identified suspect. Exa speaker's voice in the first sample. Th second sample group was recorded fro cation process will go well, and vice methods which are used in handling c well as a simulated speech signal were BACKGROUND The development of forensic speaker identif nesia was started from when speech signal s legal evidence in the court of corruption case as legal evidence to the court, the speech should be first examined by the expert. The pr signal samples analysis, or known as Speak (SV) system, is a system that identify the sub known (UK) sample by comparing it with sample. K sample is speech signal recording suspected person with the known identity, an speech signal recording taken from suspecte For example, the UK sample is taken from tel sation recording. Both of the samples (K and the Voice Comparison Standards that came Board of Recorded Evidence, a board of the lege of Forensic Examiners [1]. The legal ev proceed from SV analysis is depend on the m sis that used in the SV system. ANALYSIS METHOD OF SPEAKER VERIFICATION The method of analysis that used in SV syste legal evidence from speech signal samples Court Justice, consist of four main steps. Th pairing, tagging, acoustic features extraction analysis as can be seen in Figure 1. Proceedings of 20 th International Congress on Ac 23-27 August 2010 sic speaker identification: ience in Indonesians court ono, M.I. Mandasari, and Suprijanto hysics, Insitut Teknologi Bandung, Jl. Ganesa 10, Bandung, IND jsarwono@tf.itb.ac.id munication can be accepted as one of legal evidence in Indone e court was a case of corruption handled by the Indonesian Cor he speech evidence to court, such speech evidence should be miner to verify the identity of the speaker. Two groups of speec epted voice belongs to unidentified speaker, while the second aminer serves to compare the two sample groups and confirm his examination process is strongly influenced by processes that om the suspect. If the defendant cooperated with investigators, t e versa. This paper shows the development of Forensic Speak corruption cases in Indonesian court. Results of identification u e also discussed. fication in Indo- samples used as es. Before taken signal samples rocess or speech ker Verification bject of the Un- the Known (K) g taken from the nd UK sample is ed conversation. lephone conver- d UK) based on from American e American Col- vidence which is method of analy- em to provide as s in Indonesian he four steps are n, and statistical On the pairing step, words from the K the same words of the UK sample in o sis more efficient. Standard number commonly used is ranging from 10 un times higher depending on desirable step, a word of the speech signal sam syllables, because the speech signal s be used in the statistical analysis step. Figure 1. The four steps in the analys that used in Indonesia The SV system use Praat software f extraction step, since Praat is a freew analysis and manipulate tools for spe tion is using an autocorrelation meth tion is using the LPC-Burg method. tem for producing legal evidence in I only pitch and formant as the acoustic bandwidth formant can also be used statistical analysis. Statistical analysis was carried out to of acoustic features set from the coustics, ICA 2010 0, Sydney, Australia DONESIA esians court. The rruption Eradica- e examined by a ch samples were d is the voice re- m the identity of t occur while the then the identifi- ker Identification using real data as K sample were paired to order to make SV analy- r of pairing words that ntil 20 words, but some- e result. In the tagging mples were divided into segment per syllable will . sis method of SV system an court. for the acoustic features ware that can be used as eech signal. Pitch extrac- hod, and formant extrac- Until now, the SV sys- Indonesia Court still use c features. It is however, in some special cases of o make statistic features e K and UK samples.