Modernizing Open-Set Speech Language Identification

TL;DR

This paper explores adapting modern closed-set speech language identification methods to open-set scenarios, enabling rejection of unknown languages through thresholding and verification techniques, and compares two neural network architectures.

Contribution

It introduces modifications to CRNN and TDNN models for open-set language identification and evaluates two out-of-set detection strategies.

Findings

CRNN and TDNN models show different performance profiles.

Threshold-based and verification-based detection methods are compared.

Enhanced feature embeddings improve open-set detection accuracy.

Abstract

While most modern speech Language Identification methods are closed-set, we want to see if they can be modified and adapted for the open-set problem. When switching to the open-set problem, the solution gains the ability to reject an audio input when it fails to match any of our known language options. We tackle the open-set task by adapting two modern-day state-of-the-art approaches to closed-set language identification: the first using a CRNN with attention and the second using a TDNN. In addition to enhancing our input feature embeddings using MFCCs, log spectral features, and pitch, we will be attempting two approaches to out-of-set language detection: one using thresholds, and the other essentially performing a verification task. We will compare both the performance of the TDNN and the CRNN, as well as our detection approaches.