A phonetically aware system for speech activity detection
Speech activity detection (SAD) is an essential component of most speech processing tasks and greatly influences the performance of the systems. Noise and channel distortions remain a challenge for SAD systems. In this paper, we focus on a dataset of highly degraded signals, developed under the DARP...
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | CONF |
| Materias: | |
| Acceso en línea: | http://hdl.handle.net/20.500.12110/paper_15206149_v2016-May_n_p5710_Ferrer |
| Aporte de: |
| id |
todo:paper_15206149_v2016-May_n_p5710_Ferrer |
|---|---|
| record_format |
dspace |
| spelling |
todo:paper_15206149_v2016-May_n_p5710_Ferrer2023-10-03T16:20:33Z A phonetically aware system for speech activity detection Ferrer, L. Graciarena, M. Mitra, V. The Institute of Electrical and Electronics Engineers Signal Processing Society bottleneck features deep neural networks degraded channels Speech activity detection Speech activity detection (SAD) is an essential component of most speech processing tasks and greatly influences the performance of the systems. Noise and channel distortions remain a challenge for SAD systems. In this paper, we focus on a dataset of highly degraded signals, developed under the DARPA Robust Automatic Transcription of Speech (RATS) program. On this challenging data, the best-performing systems are those based on deep neural networks (DNN) trained to predict speech/non-speech posteriors for each frame. We propose a novel two-stage approach to SAD that attempts to model phonetic information in the signal more explicitly than in current systems. In the first stage, a bottleneck DNN is trained to predict posteriors for senones. The activations at the bottleneck layer are then used as input to a second DNN, trained to predict the speech/non-speech posteriors. We test performance on two datasets, with matched and mismatched channels compared to those in the training data. On the matched channels, the proposed approach leads to gains of approximately 35% relative to our best single-stage DNN SAD system. On mismatched channels, the proposed system obtains comparable performance to our baseline, indicating more work needs to be done to improve robustness to mismatched data. © 2016 IEEE. CONF info:eu-repo/semantics/openAccess http://creativecommons.org/licenses/by/2.5/ar http://hdl.handle.net/20.500.12110/paper_15206149_v2016-May_n_p5710_Ferrer |
| institution |
Universidad de Buenos Aires |
| institution_str |
I-28 |
| repository_str |
R-134 |
| collection |
Biblioteca Digital - Facultad de Ciencias Exactas y Naturales (UBA) |
| topic |
bottleneck features deep neural networks degraded channels Speech activity detection |
| spellingShingle |
bottleneck features deep neural networks degraded channels Speech activity detection Ferrer, L. Graciarena, M. Mitra, V. The Institute of Electrical and Electronics Engineers Signal Processing Society A phonetically aware system for speech activity detection |
| topic_facet |
bottleneck features deep neural networks degraded channels Speech activity detection |
| description |
Speech activity detection (SAD) is an essential component of most speech processing tasks and greatly influences the performance of the systems. Noise and channel distortions remain a challenge for SAD systems. In this paper, we focus on a dataset of highly degraded signals, developed under the DARPA Robust Automatic Transcription of Speech (RATS) program. On this challenging data, the best-performing systems are those based on deep neural networks (DNN) trained to predict speech/non-speech posteriors for each frame. We propose a novel two-stage approach to SAD that attempts to model phonetic information in the signal more explicitly than in current systems. In the first stage, a bottleneck DNN is trained to predict posteriors for senones. The activations at the bottleneck layer are then used as input to a second DNN, trained to predict the speech/non-speech posteriors. We test performance on two datasets, with matched and mismatched channels compared to those in the training data. On the matched channels, the proposed approach leads to gains of approximately 35% relative to our best single-stage DNN SAD system. On mismatched channels, the proposed system obtains comparable performance to our baseline, indicating more work needs to be done to improve robustness to mismatched data. © 2016 IEEE. |
| format |
CONF |
| author |
Ferrer, L. Graciarena, M. Mitra, V. The Institute of Electrical and Electronics Engineers Signal Processing Society |
| author_facet |
Ferrer, L. Graciarena, M. Mitra, V. The Institute of Electrical and Electronics Engineers Signal Processing Society |
| author_sort |
Ferrer, L. |
| title |
A phonetically aware system for speech activity detection |
| title_short |
A phonetically aware system for speech activity detection |
| title_full |
A phonetically aware system for speech activity detection |
| title_fullStr |
A phonetically aware system for speech activity detection |
| title_full_unstemmed |
A phonetically aware system for speech activity detection |
| title_sort |
phonetically aware system for speech activity detection |
| url |
http://hdl.handle.net/20.500.12110/paper_15206149_v2016-May_n_p5710_Ferrer |
| work_keys_str_mv |
AT ferrerl aphoneticallyawaresystemforspeechactivitydetection AT graciarenam aphoneticallyawaresystemforspeechactivitydetection AT mitrav aphoneticallyawaresystemforspeechactivitydetection AT theinstituteofelectricalandelectronicsengineerssignalprocessingsociety aphoneticallyawaresystemforspeechactivitydetection AT ferrerl phoneticallyawaresystemforspeechactivitydetection AT graciarenam phoneticallyawaresystemforspeechactivitydetection AT mitrav phoneticallyawaresystemforspeechactivitydetection AT theinstituteofelectricalandelectronicsengineerssignalprocessingsociety phoneticallyawaresystemforspeechactivitydetection |
| _version_ |
1782030148454842368 |