搜索资源列表
Natural gradient ML or nonlinear decorrelation alg
- 极小边际熵等价于叉四阶累积量的平方和最小。通过迭代使四阶累积矩阵对角化,实现交叉四阶累积量的平方和的极小化。他是语音识别的重要预处理算法-minimum entropy equivalent to the marginal four bands fork cumulative amount of square and smallest. Through iterative four bands so that the cumulative matrix diagonalization, four
audacity-src-1.2.3.tar
- Audacity: A Free, Cross-Platform Digital Audio Editor
VCandMATLAB
- Based on short-term energy detection and short-term cross zero rates detection in speech reorganization,the paper presents two-threshold endpoint detection.In addition,an accurate speech segmentation algorithm is achieved with the wavelet transfo
imskpe-1.0beta7-win32-full
- klatt共振峰语音合成器,可以修改模型参数达到修改合成语音音质的目的。该程序采用gtk开发,可跨平台使用-Formant voice synthesizer, model parameters can be amended to meet the synthetic voice quality purposes. The program uses gtk development, the use of cross-platform
g729lib
- g729编码程序,交叉编译成.a库文件,可以在嵌入式下使用,含测试程序-g729 coding process, cross-compiled. a library file that can be used in embedded, with test program
zcr
- 同一种思想,三种实现方法,求语音信号的短时过零率,并相互验证。运行无误-three roads to the ZCR(zero-cross rate) of speech signal according to its defination.
pinyin_python
- 能将任一分过词的文章,进行去重、排序,转换为拼音、将拼音转换为音素。可用于汉语语音识别前的语料准备。代码已在python 2.7上运行通过。-Able to any one point of the cross-word article, de-emphasis, sort, convert Pinyin Pinyin conversion to phonemes. Can be used for the corpus preparation before the Chinese speech
wav_FFT_demo
- 快速傅氏变换(FFT),允许用户查看的音频信号的频谱内容。 FFT这里给出的代码是由唐交叉出现,他的主页,随后被撤下。而不是解释的FFT的数学理论,我将试图解释它的用处,因为它涉及到的音频信号。 FFT允许用户获得的音频信号的频谱构成,获得它的各种频率的分贝,或获得其各种频率的强度。光谱观众(在上面的图片所示),均衡器,或VU表可以使用FFT,以显示其结果。它们之间的差值,然后取决于一对夫妇方程采取的FFT的实部和虚部的组件,并返回的强度或分贝水平被用于在绘制结果中的一个。下面的代
Extraction-of-periodic-signal-noise
- 提取淹没在噪声中的周期信号要求自相关(以判断周期)和互相关(以恢复信号自身)-Extraction drowned in the noise requirements of a periodic signal autocorrelation (to determine the cycle) and cross-correlation (to restore the signal itself)
speech-analysis
- 对语音进行分析,包括时域分析(包括能量、过零率、互相关函数)和频域分析(包括fft变换、倒谱、LPC)-Speech analysis, including time domain analysis (including energy, zero-crossing rate, the cross-correlation function) and frequency domain analysis (including fft transform, cepstrum, LPC)
julius-4.3.1.tar
- Julius 是一种高性能,两通大词汇量连续语音识别(LVCSR)语音相关的研究和开发的解码器软件。基于字的N-gram和上下文相关的HMM模型,它可以进行几乎实时实时解码目前大多数电脑在60K字听写任务。完全纳入,如树的N-gram词汇,保,跨词的上下文依赖处理,包围梁搜索,高斯修剪,高斯的选择,除了搜索效率等各大搜索技术,它也是模块化小心从模型结构独立,如共享状态triphones的和并列混合模型与任意数量的混合物,州或手机,支持各种HMM的类型。采用标准格式,以配合HTK的,债务工具中央结