Earwitness Testimony: Can You ID a Voice?

Imagine the scene: You are standing in a bank line or walking down a dimly lit alley. Suddenly, a chaotic event unfolds. A robbery. An assault. You dive for cover, eyes squeezed shut in fear, or perhaps the perpetrator wears a mask. You never see their face. But you hear them. You hear them shout commands, make threats, or speak to an accomplice.

Days later, the police ask you: “If you heard that voice again, would you recognize it?”

Most of us would instinctively say yes. We trust our ears. We recognize our mothers on the phone after a single syllable; we identify famous actors in animated movies without seeing the credits. However, forensic linguistics and psychological research tell a different, more troubling story. While eyewitness testimony has long been scrutinized for its unreliability, earwitness testimony—identifying a suspect by voice alone—is notoriously more fragile, yet it continues to play a pivotal role in criminal justice.

The Spectrum of Reliability: Familiarity vs. Strangers

To understand why voice identification is difficult, we must distinguish between identifying a familiar speaker and an unfamiliar one. As humans, we are linguistic experts regarding our “inner circle.” If your best friend calls you from a different number and says “Hello”, your brain instantly matches the pitch, timbre, and prosody to a stored mental model. This is high-reliability identification.

However, crimes are rarely committed by our best friends. They are usually committed by strangers. When we hear a stranger’s voice, our brain lacks a pre-existing “voice print.” We are forced to rely on short-term acoustic memory.

Research suggests that our memory for voices decays rapidly—much faster than our memory for faces. In psychological studies, accuracy rates for voice identification drop significantly after just a few hours. If a witness is asked to identify a voice a week after the crime, the likelihood of a false identification skyrockets. The brain remembers the message (the semantics) much better than the medium (the specific acoustic qualities of the speaker).

The Forensic Linguistics of the “Voice Lineup”

When visual evidence is absent, police may conduct a “voice parade” or lineup. Just as a visual lineup places a suspect among several “fillers” (lookalikes), a voice lineup plays a recording of the suspect alongside recordings of people with similar vocal characteristics.

From a linguistic perspective, constructing a fair voice lineup is a nightmare. To create a valid test, forensic linguists must control for numerous variables:

• Accent and Dialect: If the perpetrator had a Boston accent, all fillers must have a Boston accent. If the suspect is the only one with the matching dialect, the witness will pick them out not because they recognize the voice, but because they recognize the category.
• Pitch and Timbre: The fundamental frequency of the voices must be similar.
• Recording Quality: This is a common pitfall. If the suspect is recorded in a sterile police interrogation room, but the fillers are recorded on handheld recorders with background noise, the witness may unconsciously choose the “cleanest” recording.
• Utterance Length: Everyone must say the same thing for the same duration.

Even when these controls are in place, the error rate remains high. Unlike a face, which allows us to scan features simultaneously (holistic processing), voice is temporal. We have to listen to sample A, remember it, listen to sample B, compare it to the memory of A, and so on. By the time we get to sample E, our memory of sample A has degraded.

The Phonological Loop and the “Telephone” Effect

Why are we so bad at this? Part of the answer lies in how we process language. When we listen to speech, our brains prioritize meaning over sound. We are biologically wired to decode syntax and semantics to understand the threat or the instruction.

Unless you are a trained phonetician, you likely aren’t mentally cataloging the speaker’s vowel shifts, glottal stops, or vocal fry during a robbery. You are focusing on the content: “Put the money in the bag.”

Furthermore, external factors can distort perception. This is often called “channel mismatch.” If you heard the criminal screaming in an echolic bank lobby, but you are asked to identify a suspect speaking calmly in a soundproof room, the acoustic features change entirely. Stress alters the vocal cords, raising pitch and changing speed. A scream does not sound like a whisper, and a shout does not sound like a conversational tone, even when they come from the same throat.

The Lindbergh Case: A Historical Warning

One of the most famous examples of controversial earwitness testimony is the kidnapping of Charles Lindbergh’s baby in 1932. Years after the crime, Lindbergh identified the voice of Bruno Richard Hauptmann as the man he heard shouting in a cemetery nearly three years prior. Lindbergh stated, “That is the voice.”

From a modern forensic linguistic standpoint, this is terrifying. The idea that a human can retain a specific, unfamiliar voice print for three years after hearing only two words (“Hey, Doctor”) is scientifically improbable. Yet, the testimony helped send Hauptmann to the electric chair. Today, such confidence after such a long delay would be vigorously challenged by defense experts.

Linguistic Profiling and Bias

Perhaps the most insidious aspect of earwitness testimony is the intrusion of bias—what linguist John Baugh terms “linguistic profiling.”

When we hear a voice without seeing a face, we immediately construct a mental image of the speaker based on stereotypes regarding dialect, sociolect (social class markers), and gender. If a witness believes a crime was committed by a specific demographic, they are more likely to misidentify a voice that fits their customized stereotype of that demographic.

For example, if a witness hears a structurally ambiguous accent but perceives the speaker as “threatening”, they may mentally categorize the voice into a marginalized group due to social conditioning. When presented with a lineup, they may select the voice that sounds “most stereotypical”, rather than the voice they actually heard.

Can We Trust Our Ears?

This is not to say that voice identification is useless. It can be a powerful corroborative tool. However, forensic linguists argue that it should rarely be used as the sole evidence for conviction.

Technology is attempting to bridge the gap. Forensic voice comparison using spectrograms (visual representations of sound waves) and semi-automatic recognizer systems is becoming more common. These tools analyze the physics of the voice—formants, frequencies, and harmonics—stripping away human memory fallibility. But even algorithms struggle with the “mismatch” problem of high-stress shouting versus calm speech.

For language learners and enthusiasts, the takeaway is a newfound respect for the complexity of human speech. Our voices are as unique as our fingerprints, comprised of physiology, learned accents, emotional states, and social mimickry. But unlike a fingerprint, a voice is fluid, changing from moment to moment. While we may feel certain we could identify a stranger’s voice, the science suggests that when the eyes are closed, the ears are easily deceived.