AES Fall Online Convention

In-room loudspeaker equalization requires a significant amount of microphone positions in order to characterize the sound field in the room. This can be a cumbersome task for the user. This paper proposes the use of artificial intelligence to automatically estimate and equalize, without user interaction, the in-room response. To learn the relationship between loudspeaker near-field response and total sound power, or energy average over the listening area, a neural network was trained using room measurement data. Loudspeaker near-field SPL at discrete frequencies was the input data to the neural network. The approach has been tested in a subwoofer, a full-range loudspeaker, and a TV. Results showed that the in-room sound field can be estimated within 1--2 dB average standard deviation.

Traditional room-equalization involves exciting one loudspeaker at a time and deconvolving the loudspeaker-room response from the recording. As the number of loudspeakers and positions increase, the time required to measure loudspeaker-room responses will increase. In this paper, we present a technique to deconvolve impulse responses after exciting all loudspeakers at the same time. The stimuli are shifted relative to a base-stimuli and are optionally pre-processed with arbitrary filters to create specific sounding signals. The stimuli shift ensures capture of the low-frequency reverberation tail after deconvolution. Various deconvolution techniques including correlation-based, and adaptive filter-based, are presented. The performance is characterized in terms of plots and objective metrics using responses from the Multichannel Acoustic Reverberation Dataset (MARDY) dataset.

Several key studios in Nashville, TN served as the focus for the creation of the recorded music experience known as “the Nashville Sound.” Recordings were notable for their songwriting style, musical arrangement, and the nature of the technical processes employed, including the specific recording spaces themselves. Three historically significant studios were selected as representative of this era. This study reviewed the historical background of the studios and investigated whether there may be similarities in these studios’ acoustical properties that resulted in a particular recording approach within these environments. Standard acoustic measurements were obtained and analysed in each of these three recording spaces.

It is known that musicians tend to adjust their performance to the acoustical properties of the hall as they perceive. In a large reverberant hall, for example, they may play staccato notes even shorter than they would in a less reverberant hall to make the music more clearly understandable by the audience. In this study, four singers were invited to sing two (slow and fast) pieces of music in three venues, of which the reverberation times were 0.3, 1.8, and 3.4 seconds. Singers were surveyed with questions regarding the tempo, intonation, resonance and diction of their performance in each venue. Also, the singing voice was recorded by using a headset microphone and analyzed to relate the audio features to the characteristics of the venues. The results showed that the singers’ perception of the vocal resonance was significantly related to the venue (p=0.024), and so were the average sound level and the dynamic range of the sound level (p=0.040 for both dependent variables), which could partly be explained in relation to the reverberation time.