端到端的功能融合,共同优化语音增强和自动语音识别
Mohamed Medani1, Nasir Saleem2, Fethi Fkih3
1Applied College of Muhayel Aseer, King Khalid University, Abha, 62529, Saudi Arabia.
Scientific reports
|July 2, 2025
概括
这项研究介绍了一种新的语音增强 (SE) 模型,该模型可以动态地融合增强和杂的语音特征. 这种方法显著提高了语音质量,并减少了实时自动语音识别 (ASR) 系统中的错误.
科学领域:
- 信号处理 信号处理
- 人工智能的人工智能
- 机器学习 机器学习
背景情况:
- 实时语音增强 (SE) 和自动语音识别 (ASR) 对于清晰的沟通和准确的转录至关重要.
- 传统的SE方法可能会扭曲语音,对下游ASR性能产生负面影响.
- 现有的SE-ASR联合模型通常只使用增强功能,可能会丢失有价值的信息.
研究的目的:
- 开发一个可以抑制噪音的SE网络,同时最大限度地减少语音扭曲.
- 提出一种动态融合方法,整合增强和原始噪音语音功能,以改善ASR.
- 为强大的端到端ASR创建一个联合培训框架.
主要方法:
- 一个注意力编解码模型,具有SE的因果注意力机制.
- 在SE网络中,使用基于注意力的修正线性单元 (AReLU) 的修改过的Gated Recurrent Unit (GRU).
- 一个基于GRU的融合网络,结合了增强和原始噪声特征,输入ASR系统.
主要成果:
- 拟议的SE模型在匹配和不匹配的条件下,与基线相比,实现了优越的语音质量,可理解性和噪声抑制.
- 在匹配的条件下,在STOI (19.81%) 和PESQ (28.97%) 中观察到显著的改善.
- 联合培训框架减少了ASR的字符错误率 (CER) 从32.99%降至13.52%.
结论:
- 动态融合方法有效地减轻了语音扭曲,并从噪音信号中保存了关键细节.
- 拟议的SE和联合培训框架大大提高了实时ASR系统的稳定性和准确性.
- 这种综合方法为具有挑战性的声学环境提供了有希望的解决方案.
更多相关视频
相关概念视频
01:22Amplifying Signals via Enzymatic Cascade
9.3K
When a ligand binds to a cell-surface receptor, the receptor's intracellular domain changes shape, which may either activate its enzyme function or allow its binding to other molecules. The initial signal is amplified by most signal transduction pathways. This means that a single ligand molecule can activate multiple molecules of a downstream target. Proteins that relay a signal are most commonly phosphorylated at one or more sites, activating or inactivating the protein. Kinases catalyze...
9.3K
02:07Improving Translational Accuracy
11.9K
Base complementarity between the three base pairs of mRNA codon and the tRNA anticodon is not a failsafe mechanism. Inaccuracies can range from a single mismatch to no correct base pairing at all. The free energy difference between the correct and nearly correct base pairs can be as small as 3 kcal/ mol. With complementarity being the only proofreading step, the estimated error frequency would be one wrong amino acid in every 100 amino acids incorporated. However, error frequencies observed in...
11.9K
00:56Extraction: Advanced Methods
547
Metal ions can be separated from one another by complexation with organic ligands–the chelating agent– to form uncharged chelates. Here, the chelating agent must contain hydrophobic groups and behave as a weak acid, losing a proton to bind with the metal. Since most organic ligands used in this process are insoluble or undergo oxidation in the aqueous phase, the chelating agent is initially added to the organic phase and extracted into the aqueous phase. The metal-ligand complex is...
547
01:12Double Resonance Techniques: Overview
306
Double resonance techniques in Nuclear Magnetic Resonance (NMR) spectroscopy involve the simultaneous application of two different frequencies or radiofrequency pulses to manipulate and observe two distinct nuclear spins. One important application of double resonance is spin decoupling, which selectively suppresses coupling with one type of nucleus while observing the NMR signal from another nucleus, simplifying the spectrum and enhancing resolution.
Spin decoupling is usually achieved by...
Spin decoupling is usually achieved by...
306
01:35Pre-mRNA Processing: Modification of pre-mRNA Ends
10.0K
In eukaryotic cells, transcripts made by RNA polymerase are modified and processed before exiting the nucleus. Unprocessed RNA is called precursor mRNA or pre-mRNA to distinguish it from mature mRNA.
Once about 20-40 ribonucleotides have been joined together by RNA polymerase, a group of enzymes adds a cap to the 5' end of the growing transcript. In this process, a 5' phosphate is replaced by modified guanosine that has a methyl group attached (7-methyl guanosine). This 5' cap helps...
Once about 20-40 ribonucleotides have been joined together by RNA polymerase, a group of enzymes adds a cap to the 5' end of the growing transcript. In this process, a 5' phosphate is replaced by modified guanosine that has a methyl group attached (7-methyl guanosine). This 5' cap helps...
10.0K
01:26Linear Approximation in Frequency Domain
137
Linear systems are characterized by two main properties: superposition and homogeneity. Superposition allows the response to multiple inputs to be the sum of the responses to each individual input. Homogeneity ensures that scaling an input by a scalar results in the response being scaled by the same scalar.
In contrast, nonlinear systems do not inherently possess these properties. However, for small deviations around an operating point, a nonlinear system can often be approximated as linear....
In contrast, nonlinear systems do not inherently possess these properties. However, for small deviations around an operating point, a nonlinear system can often be approximated as linear....
137