梅森-阿尔伯塔语音分割器:一种基于深度神经网络和插曲的强制对齐系统
Matthew C Kelley1, Scott James Perry2, Benjamin V Tucker2,3
1Department of English, Linguistics Program, George Mason University 3298 , Fairfax, VA, USA.
Phonetica
|September 9, 2024
概括
我们开发了一种新的神经网络强制对齐系统,Mason-Alberta Phonetic Segmenter (MAPS),可以显著改善语音段边界检测. MAPS的性能优于当前最先进的系统,为大语音体分析提供更精确的结果.
科学领域:
- 计算语言学 计算语言学
- 语音处理 语音处理
- 人工智能的人工智能
背景情况:
- 强制对齐系统对于细分语音数据至关重要,使得大规模的语料库分析成为可能.
- 现有系统在边界精度和声学建模方法上往往存在局限性.
研究的目的:
- 介绍梅森-阿尔伯塔语音分段器 (MAPS),一种基于神经网络的新型强制对齐系统.
- 评估两个关键的改进:将声学模型视为标记器,并采用插值技术来确定精确的边界.
主要方法:
- 开发了一个用于强制对齐的神经网络架构.
- 实现了一个声学模型作为标记器,承认重叠的语音段.
- 使用插值技术来完善边界检测超出典型的10毫秒限制.
主要成果:
- 所有MAPS配置的性能都显著超过了蒙特利尔强制对齐器在10毫秒的宽容度.
- 在边界定位准确度方面实现了高达28.13%的相对性能提升.
- 观察到蒙特利尔强制对齐器仅在30毫秒的容忍度下略高于MAPS的性能.
结论:
- 与最先进的系统相比,MAPS在精确的语音边界检测方面表现出卓越的性能.
- 这项研究强调了声学模型培训目标与语音相似性之间的潜在差异,并建议了未来的研究方向.
- 重新思考声学建模和输出目标可能是必要的,以进一步推进强制对齐.
更多相关视频
相关概念视频
One-Degree-of-Freedom System
954
In mechanical engineering, one-degree-of-freedom systems form the basis of a wide range of electrical and mechanical components. Using these models, engineers can predict the behavior of various parts in a larger system, which gives them insight into how different forces interact with each other.
A one-degree-of-freedom system is defined by an independent variable that determines its state and behavior. One example of a one-degree-of-freedom system is a simple harmonic oscillator, such as a...
A one-degree-of-freedom system is defined by an independent variable that determines its state and behavior. One example of a one-degree-of-freedom system is a simple harmonic oscillator, such as a...
954
Reconstruction of Signal using Interpolation
878
Signal processing techniques are essential for accurately converting continuous signals to digital formats and vice versa. When a continuous signal is sampled with a period T, the resulting sampled signal exhibits replicas of the original spectrum in the frequency domain, spaced at intervals equal to the sampling frequency. To handle this sampled signal, a zero-order hold method can be applied, which creates a piecewise constant signal by retaining each sample's value until the next...
878


