Peer-reviewed research

Speech recognition by machines and humans

Richard P. Lippmann, Speech Communication, vol. 22, pp. 1-15, 1997

Read the original

Read on October 17, 2026

What it is: A review from MIT Lincoln Laboratory that compared machine and human error rates across six speech corpora, from isolated words to spontaneous telephone conversation. It is the classic benchmark of how far machines lagged humans in the 1990s.
How much weight it carries: Peer-reviewed and widely cited. The corpora are old and the machine systems are 1990s HMM recognizers, so the numbers show the historical gap, not today's accuracy. Human and machine results come from different tests, so compare the two within each row, not across rows.

Facts taken from this source

In a 1997 review of noise experiments, a machine with noise adaptation reached 12.8 percent error at a signal-to-noise ratio of 10 dB, more than ten times the 1 percent human error rate.
“Even with noise adaptation, however, the 12.8% machine error rate shown in Fig. 5 at an SNR of 10 dB is more than an order of magnitude greater than the 1% human error rate.” (Section 4 (noise experiments, Fig. 5))
In a 1997 review of six speech corpora, machine error rates were often more than ten times higher than human error rates on quiet, wideband, read speech.
“Error rates of machines are often more than an order of magnitude greater than those of humans for quiet, wideband, read speech.” (Abstract)
Lippmann's 1997 paper reviews past work comparing modern speech recognition systems with humans, to see how far the technology has moved toward human-like performance.
“This paper reviews past work comparing modern speech recognition systems and humans to determine how far recent dramatic advances in technology have progressed towards the goal of human-like performance.” (Abstract)
In the same 1997 review, human transcribers had an error rate of 4 percent on spontaneous telephone conversation, more than ten times lower than the 43 percent of the best machine recognizer.
“Once again, the 4% error rate of humans is more than an order of magnitude less than the 43% error rate” (Section 4.4 (Switchboard))
In a 1997 review, an HMM recognizer had a word error rate of 43 percent on the spontaneous telephone conversations of the Switchboard corpus.
“The 43% error rate of the HMM recognizer is extremely high compared to the much lower machine rates” (Section 4.4 (Switchboard))

Pages that rest on this source

Peer-reviewed research

Speech recognition by machines and humans

Richard P. Lippmann, Speech Communication, vol. 22, pp. 1-15, 1997

Read the original

Read on October 17, 2026

What it is: A review from MIT Lincoln Laboratory that compared machine and human error rates across six speech corpora, from isolated words to spontaneous telephone conversation. It is the classic benchmark of how far machines lagged humans in the 1990s.
How much weight it carries: Peer-reviewed and widely cited. The corpora are old and the machine systems are 1990s HMM recognizers, so the numbers show the historical gap, not today's accuracy. Human and machine results come from different tests, so compare the two within each row, not across rows.

Facts taken from this source

In a 1997 review of noise experiments, a machine with noise adaptation reached 12.8 percent error at a signal-to-noise ratio of 10 dB, more than ten times the 1 percent human error rate.
“Even with noise adaptation, however, the 12.8% machine error rate shown in Fig. 5 at an SNR of 10 dB is more than an order of magnitude greater than the 1% human error rate.” (Section 4 (noise experiments, Fig. 5))
In a 1997 review of six speech corpora, machine error rates were often more than ten times higher than human error rates on quiet, wideband, read speech.
“Error rates of machines are often more than an order of magnitude greater than those of humans for quiet, wideband, read speech.” (Abstract)
Lippmann's 1997 paper reviews past work comparing modern speech recognition systems with humans, to see how far the technology has moved toward human-like performance.
“This paper reviews past work comparing modern speech recognition systems and humans to determine how far recent dramatic advances in technology have progressed towards the goal of human-like performance.” (Abstract)
In the same 1997 review, human transcribers had an error rate of 4 percent on spontaneous telephone conversation, more than ten times lower than the 43 percent of the best machine recognizer.
“Once again, the 4% error rate of humans is more than an order of magnitude less than the 43% error rate” (Section 4.4 (Switchboard))
In a 1997 review, an HMM recognizer had a word error rate of 43 percent on the spontaneous telephone conversations of the Switchboard corpus.
“The 43% error rate of the HMM recognizer is extremely high compared to the much lower machine rates” (Section 4.4 (Switchboard))