没有合适的资源?快使用搜索试试~ 我知道了~
首页长短帧关联谐波模型提升单通道语音分离效果
长短帧关联谐波模型提升单通道语音分离效果
1 下载量 167 浏览量
更新于2024-08-27
收藏 501KB PDF 举报
本文主要探讨了"基于长短帧相关谐波模型的单通道语音分离"这一技术在音频处理领域的应用。传统的谐波模型在音乐源分离中表现出色,因为不同音乐源间的谐波峰值具有显著差异,这使得它们能有效地识别和分离。然而,在处理语音信号时,由于短时间窗口的存在,会不可避免地导致频域中的谐波重叠问题,影响到分离的精度。 针对这一挑战,研究人员提出了长-短帧相关谐波(LSAH)模型。长帧提供更高的谐波分辨率,能够更精确地捕捉语音信号中的高频成分,而短帧则保证了信号的瞬时稳定性,有助于减少时间窗口对频率分析的影响。这种结合利用了两种帧长度的优势,有助于提高多音高估计(multi-pitch estimation)的准确性。 LSAH模型的关键在于自相关方法的应用,这种方法既简单又高效,能够准确地确定音高的突出部分。通过这种方法,模型可以判断混音状态并估算潜在的其他音高候选,从而实现有效的语音分离。相比于传统的短时谐波模型,LSAH模型在保持高谐波分辨率的同时,还能更好地处理清音段,这是许多现有技术难以处理的一个优点。 作者们通过在30组混合信号上进行实验,验证了LSAH模型在信噪比(SNR)和主观听感质量方面的优势。结果表明,LSAH算法在处理单通道语音分离任务时,不仅提高了分离效果,还提供了更佳的用户体验,这使其在实际应用中具有很高的实用价值。 本文的核心贡献在于提出了一种创新的音频处理策略,利用长短帧相结合的方式,有效地解决了单通道语音分离中的谐波重叠问题,提升了语音信号的分离质量和稳定性,对于音频处理领域,特别是语音增强、噪声抑制等方面的研究具有重要意义。
资源详情
资源推荐
Q. Huang, D. Wang / Digital Signal Processing 21 (2011) 497–507 499
Fig. 1. Spectrum and the peaks of the long frame and the short frame signal. (a) The mixed spectrum and peaks of the long frame and the short frame signal. (b) The
spectrum and peaks of the long frame signal. (c) The spectrum and peaks of the short frame signal.
and harmonic peaks of the long frame and the short frame sig-
nal. From the short frame (c) we can see that in the frequency
range [600–700] Hz and [750–850] Hz, the harmonics of the two
sources deviate from their original position and a new harmonic
peak is formed between them. In the long frame (b) they remain
at the original positions. At about 250 Hz, the harmonic from
one source is covered by that of the competitive source in the
short frame (c), while the harmonic peaks remain there in the
long frame (b). From (a) it can be seen that harmonics from the
two types of frames coincide with each other in frequencies of
the non-overlapping harmonic positions. However many false har-
monic peaks appear in the long frame spectrum due to the short
stationary feature of the speech signal. Therefore we use the LSAH
model to overcome the drawback.
The LSAH model is represented as follows:
• Long frame harmonic structure:
A
L
=
A
l
F
1
, A
l
F
2
,...,A
l
F
Rl
(6)
• Short frame harmonic structure:
A
S
=
A
sh
F
1
, A
sh
F
2
,...,A
sh
F
Rsh
(7)
• Long–short frame associated harmonic model:
A
LS
=
A
sh
F
1
, A
sh
F
2
,...,A
sh
F
Rsh
,
short frame
A
l
F
1
, A
l
F
2
,...,A
l
F
Rl
,
long frame
(8)
A
L
and
A
S
are jointly used in the separation. All the frequency
distances between the neighboring spectral peaks both in the long
frame and the short frame are used as the fundamental factors for
state judging and pitch estimation.
3. Multi-pitch tracking for the single-channel s peech mixture
Multi-pitch estimation affects the harmonic structure selection
and the clustering, and it is the kernel of the proposed algorithm.
Many methods have been proposed for multi-F
0
estimation of mu-
sic mixture and obtained good results [18,19,23,24,30]. However,
there are few effective methods contributing to speech signal. We
propose an approach to estimate the multi-pitches of speech mix-
ture with LSAH model in three stages. Firstly prominent pitch is
estimated by autocorrelation method. Then the state of the mix-
ture is judged by LSAH model. Finally the other pitch is estimated
according to the estimated state and the prominent pitch based on
LSAH model. The accuracy of pitch estimation for the other source
is improved. Fig. 2 shows the whole process of the multi-pitch es-
timation.
3.1. Estimating the prominent pitch
Prominent pitch is estimated by single speaker pitch detection
method in previous work [3,31]. We propose to use the autocor-
relation algorithm to estimate the prominent pitch. A lot of exper-
剩余10页未读,继续阅读
weixin_38740144
- 粉丝: 1
- 资源: 972
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- C++标准程序库:权威指南
- Java解惑:奇数判断误区与改进方法
- C++编程必读:20种设计模式详解与实战
- LM3S8962微控制器数据手册
- 51单片机C语言实战教程:从入门到精通
- Spring3.0权威指南:JavaEE6实战
- Win32多线程程序设计详解
- Lucene2.9.1开发全攻略:从环境配置到索引创建
- 内存虚拟硬盘技术:提升电脑速度的秘密武器
- Java操作数据库:保存与显示图片到数据库及页面
- ISO14001:2004环境管理体系要求详解
- ShopExV4.8二次开发详解
- 企业形象与产品推广一站式网站建设技术方案揭秘
- Shopex二次开发:触发器与控制器重定向技术详解
- FPGA开发实战指南:创新设计与进阶技巧
- ShopExV4.8二次开发入门:解决升级问题与功能扩展
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功