A Method for Conversational Signal-to-Noise Ratio Estimation in Real-World Sound Scenarios

Naim Mansour, Marton Marschall, Tobias May, Adam Westermann, Torsten Dau

Research output: Contribution to journalConference abstract in journalResearchpeer-review

987 Downloads (Pure)


The analysis of conversational signal-to-noise ratios (SNRs) measured in real-world scenarios can provide vital insight into people's communicative strategies and difficulties and guide development of hearing devices. However, measuring SNRs accurately and realistically is challenging in typical recording conditions, where only a mixture of sound sources is captured. This study introduces a novel method for realistic in situ SNR estimation, where the speech signal of a person in natural conversation is captured by a cheek-mounted microphone, adjusted for free-field conditions, and convolved with a measured impulse response to estimate the clean speech component at the receiver. A microphone near the receiver computes the noise-only component by applying a voice activity detector. Obtained SNR values are analyzed using in situ recordings of a real-world workspace meeting. It is shown that the temporal resolution is increased, and fluctuations in the speech level are more accurately tracked compared to a typical spectral-subtraction-derived method. The application of the proposed SNR estimation method may be valuable for compensation procedures in hearing instruments that take conversational dynamics into account.

Original languageEnglish
JournalAcoustical Society of America. Journal
Pages (from-to)1873
Publication statusPublished - 2019
Event177th Meeting of the Acoustical Society of America - The Galt House, Louisville, United States
Duration: 13 May 201917 May 2019
Conference number: 177


Conference177th Meeting of the Acoustical Society of America
LocationThe Galt House
Country/TerritoryUnited States
Internet address


Dive into the research topics of 'A Method for Conversational Signal-to-Noise Ratio Estimation in Real-World Sound Scenarios'. Together they form a unique fingerprint.

Cite this