Time-Domain Joint Training Strategies of Speech Enhancement and Intent Classification Neural Models
Robustness against background noise and reverberation is essential for many real-world speech-based applications. One way to achieve this robustness is to employ a speech enhancement front-end that, independently of the back-end, removes the environmental perturbations from the target speech signal....
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-01-01
|
| Series: | Sensors |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1424-8220/22/1/374 |
| _version_ | 1827667561774317568 |
|---|---|
| author | Mohamed Nabih Ali Daniele Falavigna Alessio Brutti |
| author_facet | Mohamed Nabih Ali Daniele Falavigna Alessio Brutti |
| author_sort | Mohamed Nabih Ali |
| collection | DOAJ |
| description | Robustness against background noise and reverberation is essential for many real-world speech-based applications. One way to achieve this robustness is to employ a speech enhancement front-end that, independently of the back-end, removes the environmental perturbations from the target speech signal. However, although the enhancement front-end typically increases the speech quality from an intelligibility perspective, it tends to introduce distortions which deteriorate the performance of subsequent processing modules. In this paper, we investigate strategies for jointly training neural models for both speech enhancement and the back-end, which optimize a combined loss function. In this way, the enhancement front-end is guided by the back-end to provide more effective enhancement. Differently from typical state-of-the-art approaches employing on spectral features or neural embeddings, we operate in the time domain, processing raw waveforms in both components. As application scenario we consider intent classification in noisy environments. In particular, the front-end speech enhancement module is based on Wave-U-Net while the intent classifier is implemented as a temporal convolutional network. Exhaustive experiments are reported on versions of the Fluent Speech Commands corpus contaminated with noises from the Microsoft Scalable Noisy Speech Dataset, shedding light and providing insight about the most promising training approaches. |
| first_indexed | 2024-03-10T03:20:03Z |
| format | Article |
| id | doaj.art-5084c273c0bb427aabe1d176ba50ec46 |
| institution | Directory Open Access Journal |
| issn | 1424-8220 |
| language | English |
| last_indexed | 2024-03-10T03:20:03Z |
| publishDate | 2022-01-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Sensors |
| spelling | doaj.art-5084c273c0bb427aabe1d176ba50ec462023-11-23T12:21:19ZengMDPI AGSensors1424-82202022-01-0122137410.3390/s22010374Time-Domain Joint Training Strategies of Speech Enhancement and Intent Classification Neural ModelsMohamed Nabih Ali0Daniele Falavigna1Alessio Brutti2Information Engineering and Computer Science School, University of Trento, 38121 Trento, ItalyFondazione Bruno Kessler, 38121 Trento, ItalyFondazione Bruno Kessler, 38121 Trento, ItalyRobustness against background noise and reverberation is essential for many real-world speech-based applications. One way to achieve this robustness is to employ a speech enhancement front-end that, independently of the back-end, removes the environmental perturbations from the target speech signal. However, although the enhancement front-end typically increases the speech quality from an intelligibility perspective, it tends to introduce distortions which deteriorate the performance of subsequent processing modules. In this paper, we investigate strategies for jointly training neural models for both speech enhancement and the back-end, which optimize a combined loss function. In this way, the enhancement front-end is guided by the back-end to provide more effective enhancement. Differently from typical state-of-the-art approaches employing on spectral features or neural embeddings, we operate in the time domain, processing raw waveforms in both components. As application scenario we consider intent classification in noisy environments. In particular, the front-end speech enhancement module is based on Wave-U-Net while the intent classifier is implemented as a temporal convolutional network. Exhaustive experiments are reported on versions of the Fluent Speech Commands corpus contaminated with noises from the Microsoft Scalable Noisy Speech Dataset, shedding light and providing insight about the most promising training approaches.https://www.mdpi.com/1424-8220/22/1/374joint trainingspeech enhancementintent classification |
| spellingShingle | Mohamed Nabih Ali Daniele Falavigna Alessio Brutti Time-Domain Joint Training Strategies of Speech Enhancement and Intent Classification Neural Models Sensors joint training speech enhancement intent classification |
| title | Time-Domain Joint Training Strategies of Speech Enhancement and Intent Classification Neural Models |
| title_full | Time-Domain Joint Training Strategies of Speech Enhancement and Intent Classification Neural Models |
| title_fullStr | Time-Domain Joint Training Strategies of Speech Enhancement and Intent Classification Neural Models |
| title_full_unstemmed | Time-Domain Joint Training Strategies of Speech Enhancement and Intent Classification Neural Models |
| title_short | Time-Domain Joint Training Strategies of Speech Enhancement and Intent Classification Neural Models |
| title_sort | time domain joint training strategies of speech enhancement and intent classification neural models |
| topic | joint training speech enhancement intent classification |
| url | https://www.mdpi.com/1424-8220/22/1/374 |
| work_keys_str_mv | AT mohamednabihali timedomainjointtrainingstrategiesofspeechenhancementandintentclassificationneuralmodels AT danielefalavigna timedomainjointtrainingstrategiesofspeechenhancementandintentclassificationneuralmodels AT alessiobrutti timedomainjointtrainingstrategiesofspeechenhancementandintentclassificationneuralmodels |