The ACLEW DiViMe: An Easy-to-use Diarization Tool

Abstract

We present "DiViMe", an open-source virtual machine aimed at packaging speech technology for real-life data, and developed in the context of the "Analyzing Children's Language Environments across the World" Project. This first release focuses on Speech Activity Detection, Speaker Diarization, and their evaluation. The present paper introduces the set of included tools and the current workflow, which is focused on making minimal assumptions regarding users' technical skills. Additionally, we show how the current DiViMe tools fare against three sets of challenging data. In a first experiment, we look at performance with samples extracted from daylong recordings gathered using the LENA TM system from English-learning children. We find that the performance of the tools currently in DiViMe is not far from that achieved by the LENA TM proprietary software. In a second experiment, we generalize to other samples of child-centered daylong files, gathered with non-LENA TM hardware from non-English-learning children, showing that performance does not degrade in this condition. Finally, we report on performance in the DiHARD 2018 Challenge Test Data. Originally conceived in the "Speech Recognition Virtual Kitchen", DiViMe is a promising platform for packaging speech technology tools for widespread re-use, with potential impact on both fundamental and applied speech and language research.

Diarization in Maximally Ecological Recordings: Data from Tsimane Children

“…Indeed, the two systems included in DiViMe performed near the bottom in a recent challenge called DiHARD, aimed at assessing diarization performance in "difficult" datasets, such as meetings and doctor-patient interviews [33]. But it is likely that the present recordings are considerably more difficult even than their "difficult" datasets, since (according to [16]) the DiViMe pipelines applied to the DiHARD standardized evaluation set led to DERs of 65-72%. The difference between these DiHARD scores in the 70's and the DERs averaging 110% obtained for the Tsimane dataset represents the additional difficulty posed by these data.…”

“…The Diarization Virtual MachinE (DiViMe for short) currently contains two tools that permit speech activity detection (i.e., detecting which portions of the recording contain some speech), and one tool for speaker diarization (i.e., attributing a speech portion to one or another speaker), which, combined, lead to two purely unsupervised pipelines yielding a segmentation of the recording into different speakers. In a recently accepted paper [16], global speech activity detection and talker diarization performance was reported.…”

Automatic word count estimation from daylong child-centered recordings in various language environments using language-independent syllabification of speech

“…However, now that the basic concept and its functionality have been validated, the next efforts should be directed toward the development and testing of a robust speaker diarization module required for speaker attribution. Although the current ACLEW virtual machine published in Le Franc et al (2018) already contains one such a tool, DiarTK (Vijayasenan & Valente, 2012), its performance was found to be lacking on child daylong data (see also DiHARD diarization challenge 10 where DiarTK scored at the bottom among all the submissions; see also Le Franc et al, 2018). In order to maintain focus on SAD and syllabifier comparisons, no separate experiments with diarization tools were included in the present report.…”

A step-by-step guide to collecting and analyzing long-format speech environment (LFSE) recordings

“…Although the there is no real open-source, languagegeneral, and population-general version of LENA, there is a relatively easy-to-use, open alternative being developed: DiViMe (Le Franc et al, 2018;ACLEW/DiViMe, 2018).…”