The 2005 AMI System for the Transcription of Speech in Meetings

Abstract

In this paper we describe the 2005 AMI system for the transcription of speech in meetings used for participation in the 2005 NIST RT evaluations. The system was designed for participation in the speech to text part of the evaluations, in particular for transcription of speech recorded with multiple distant microphones and independent headset microphones. System performance was tested on both conference room and lecture style meetings. Although input sources are processed using different front-ends, the recognition process is based on a unified system architecture. The system operates in multiple passes and makes use of state of the art technologies such as discriminative training, vocal tract length normalisation, heteroscedastic linear discriminant analysis, speaker adaptation with maximum likelihood linear regression and minimum word error rate decoding. In this paper we describe the system performance on the official development and test sets for the NIST RT05s evaluations. The system was jointly developed in less than 10 months by a multi-site team and was shown to achieve very competitive performance.

Transcribing Meetings With the AMIDA Systems

“…Table IV shows experimental results on the benefit of multiple microphones. Experiments were conducted using different acoustic data from the same meetings, including beamforming (as described in [5]) with varying numbers of microphones. All acoustic models were trained using standard maximum-likelihood estimation on the AMI corpus only and results were obtained on two different 6-hour test sets, one only with segments with overlapped speech and one without.…”

“…Hence, the approach used needs to allow for variation, even though the NIST paradigm allows systems to make explicit use of meeting room characteristics, the AMIDA systems have not done so. Instead the audio signal is enhanced by beam-forming based on time-delay-of-arrival (TDOA) that can be used with any number of microphones [5]. After noise-filtering of the audio channels the TDOA is estimated.…”

“…The transcription systems presented here are the joint effort of the AMI/AMIDA group, in a close international collaboration that participated in the RT evaluations every year since 2005 [5] with very competitive results in 2007 and 2009, on far and near-field tasks [6]. The rest of the paper is organized as follows: We first discuss data resources available for meeting processing and their properties, with special focus on the AMI corpus.…”

Probabilistic and Bottle-Neck Features for LVCSR of Meetings

“…The system is based on AMI-LVCSR system used in NIST RT'05 evaluation [13]. Here we only summarize the main features of the system: Data -The task is the recognition of meetings defined in NIST RT'05 evaluation.…”

Recognition and understanding of meetings the AMI and AMIDA projects

“…Delays and scale factors are then used in the final stage implementing superdirective beam-forming. More details can be found in [21].…”