05 / Applied AI / 2026
Speech modelling and detection.
Speech technology / VAD + GMM
01 / SCOPE
Problem and
scope.
The first task classified speech and silence using neural voice-activity detectors, testing generalisation on a separate speech corpus. The second used diagonal-covariance Gaussian mixture models for vowel classification and closed-set speaker identification. These were separate experiments with different evaluation tasks.
Individual modelling and evaluation work using supplied datasets and practical frontends. The neural VAD and GMM implementations are available in the public repository.
02 / IMPLEMENTATION
Implementation
details.
- 01Acoustic features
- 02Context / sequence
- 03Neural VAD
- 04Post-processing
- 05Cross-corpus test
MLP, recurrent and temporal-convolutional VAD.
Move from a contextual MLP to bidirectional recurrent models and dilated temporal convolutions, examining how temporal context changes detection.
Cross-corpus VAD evaluation.
Use detection-error trade-off curves and equal error rate to compare the selected systems on a different speech corpus.
GMM vowel and speaker classification.
Use diagonal-covariance Gaussian mixture models for vowels and speaker identification, comparing mixture complexity, utterance aggregation and UBM-MAP adaptation.
03 / RESULTS
Results.
Selected reported cross-corpus results over 495,486 test frames. The tuned TCN includes median post-processing. These figures are retained evaluation results, not a new benchmark run.
Context & limitations
Datasets, labels and checkpoints are excluded from the public repository. The GMM workflow needs a supplied or replacement MFCC frontend. Closed-set speaker results do not establish general-world identification performance.
Productiv / Genesys.
Team-built Rails productivity platform