Preprint

Realistic multi-microphone data simulation for distant speech recognition

arXiv preprint, 2017

Read on October 16, 2026

What it is: A preprint that compares speech recognition on real distant-microphone recordings with close-talking recordings in a room, and reports the close-talking baseline.
How much weight it carries: Preprint, not peer-reviewed at the time of reading. The experiments use a WSJ read-speech corpus in one room with a ceiling microphone array, so the numbers describe that setup and not a phone or laptop.

Facts taken from this source

In the same study, close-talking read speech recorded in a studio had a word error rate of 3.7 percent with DNN models.
“The Word Error Rate (WER%) obtained by decoding the close-talking WSJ sentences recorded in the recording studio is 3.7” (Section 4.1)
In a study that re-recorded read speech with a distant microphone, recognition performance was dramatically reduced compared with close-talking speech.
“Results clearly highlight that in the case of distant-speech input the ASR performance is dramatically reduced” (Section 4.2)
The 2017 paper says distant speech recognition still lacks robustness when non-stationary noise and acoustic reverberation are present.
“especially when adverse acoustic conditions originated by non-stationary noises and acoustic reverberation are met” (Introduction)

Preprint

arXiv preprint, 2017

Read on October 16, 2026

What it is: A preprint that compares speech recognition on real distant-microphone recordings with close-talking recordings in a room, and reports the close-talking baseline.
How much weight it carries: Preprint, not peer-reviewed at the time of reading. The experiments use a WSJ read-speech corpus in one room with a ceiling microphone array, so the numbers describe that setup and not a phone or laptop.

In the same study, close-talking read speech recorded in a studio had a word error rate of 3.7 percent with DNN models.
“The Word Error Rate (WER%) obtained by decoding the close-talking WSJ sentences recorded in the recording studio is 3.7” (Section 4.1)
In a study that re-recorded read speech with a distant microphone, recognition performance was dramatically reduced compared with close-talking speech.
“Results clearly highlight that in the case of distant-speech input the ASR performance is dramatically reduced” (Section 4.2)
The 2017 paper says distant speech recognition still lacks robustness when non-stationary noise and acoustic reverberation are present.
“especially when adverse acoustic conditions originated by non-stationary noises and acoustic reverberation are met” (Introduction)