Blind compensation of inter-channel sampling frequency mismatch with maximum likelihood estimation in STFT domain

Abstract

This paper proposes a novel blind compensation of sampling frequency mismatch for asynchronous microphone array. Digital signals simultaneously observed by different recording devices have drift of the time differences between the observation channels because of the sampling frequency mismatch among the devices. Based on the model that such the time difference is constant within each time frame, but varies proportional to the time frame index, the effect of the sampling frequency mismatch can be compensated in the short-time Fourier transform domain by the linear phase shift. By assuming the sources are motionless and stationary, a likelihood of the sampling frequency mismatch is formulated. The maximum likelihood estimation is obtained effectively by a golden section search.

Joint speaker localization and array calibration using expectation-maximization

“…Another important challenge in ad hoc networks, tightly connected to the calibration process, is the clock synchronization. Both acoustic and non-acoustic solutions were proposed to overcome this challenge [36][37][38][39][40][41][42][43][44][45]. In the current work, we assume that the nodes are perfectly synchronized, by possibly using one of those approaches.…”

“…As this cost function in (42) includes both the arrays' and speakers' positions, the search for a global minimum is a cumbersome task.…”

Multi-Talker Speech Recognition Based on Blind Source Separation with ad hoc Microphone Array Using Smartphones and Cloud Storage

“…In this case, conventional microphone array processing does not work without modification. To overcome this problem, we have developed a blind synchronization technique [21]- [23], which compensates both the difference in the time offset and the sampling frequency mismatch. In our previous work, we showed by simulative experiments that this technique is effective for improving the source separation performance.…”

“…After recording, an iPhone app that we developed auomatically transfers each recorded signal to fixed Dropbox storage. The collected signals are synchronized by blind synchronization [21]- [23], then auxiliary-function-based independent vector analysis (AuxIVA) [24] is applied for BSS. Finally, each separated signal is transcribed by ASR.…”

“…To obtain this, a blind synchronization technique [21]- [23] that we have previously developed is applied. In this method, the time offset is first compensated using the cross-correlation of the two observed signals.…”