Mysore-based AIISH has been helping the police in voice recognition in criminal cases
‘It becomes a clinching proof in cases where there is no other evidence’
MYSORE: Not many are aware that the Mysore-based All India Speech and Hearing Institute (AIISH), the only institute in South Asia which brings together the activities of research, education, public education, training and clinical services in speech and hearing on a single platform, lends technical support to agencies like the police, Crime Investigation Department (CID) and Central Bureau of Investigation (CBI) in forensic examination cases.
AIISH is perhaps the only institute after the five forensic laboratories located to conduct specialised tests to identify voice or speech of suspects in cases such as murder, kidnap, threats and so on, which sometimes emerge as important evidence in solving the cases.
The institute has state-of-the-art equipment and specialised software to recognise voice in recorded samples and suspect samples given in the form of cassettes, compact discs, micro and mini cassettes and so on to it by the investigating agencies or the courts.
The Department of Speech Language Sciences at AIISH has been doing speech recognition analysis, a significant tool in forensic studies, since 1987. It gets cases such as kidnap, threats over phone, murders, conspiracy not just from the police of the southern States, but also from Delhi, Chandigarh and other North Indian States because of its reputation and pioneering studies in the field.
Department head S.R. Savitri told The Hindu that speech recognition was usually done by perceptual analysis and acoustics analysis. In most western countries, automatic speaker identification system or character matching was employed to recognise the voice.
“But nothing can match a human brain. So we employ those methods using speech sound characters, acoustics characters, and resonance frequency as parameters for detection,” she said.
According to the expert, it was difficult to get accurately recognise the voice in the recorded samples and suspect samples. “Only once in a murder case we had got 100 per cent result in the analysis of taped telephonic conversation of the accused. The results are usually 67 per cent and above,” she said.
Since suspect voice acted as complimentary evidence in a case, it was not the only evidence to decide the case in a court of law. However, in cases where there was no evidence except the voice of the suspect person, then voice identification becomes decisive, Dr. Savitri said.
In perceptual analysis, sentences in the recorded samples and suspect samples are compared for similarities.
However, in acoustics analysis, words in the samples were separated and fed into a computer loaded with specialised software for spectrogram studies. In each word, 12 parameters were studied to prepare a percentage of the words (in spectrograph) in the samples for similarities, she said.
Nevertheless, perceptual method needs a “trained ear” that was qualified and experienced speech pathologists to recognise the voice. While interpreting words from recorded samples and suspect samples goes awry in acoustics analysis, the percentage of result will fall below 67 which is not accurate in such studies. “Therefore, the voice identification is technical and needs utmost concentration,” she said.
According to Dr. Savitri, the experts and speech pathologists in her department took at least one week or 56 hours to identify the voice in a case.
They may take even more time if the cases are complex (voice is not clear or filled with noise and so on) and when sound waves have to be converted from analog form to digital form.
“We are at present working on five cases from the southern States.
Last month, we had handled seven cases, including one from the forensic laboratory in Chandigarh. We charge Rs. 2,000 a case and take up voice identification request from counsels too,” she said.
Dr. Savitri said the analysis reports were given to the agencies concerned which were subsequently produced in the court of law as evidence.
“If we cannot recognise the voice because of difficulties, then we hand over the samples back to the police. Sometimes, we are even asked to appear before the court to certify the voice recognised by us,” she said.