| 摘要 |
Signal separation techniques based on frequency dependency are described. In one implementation, a blind signal separation process is provided that avoids the permutation problem of previous signal separation processes. In the process, two or more signal sources are provided, with each signal source having recognized frequency dependencies. The process uses these inter-frequency dependencies to more robustly separate the source signals. The process receives a set of mixed signal input signals, and samples each input signal using a rolling window process. The sampled data is transformed into the frequency domain, which provides channel inputs to the inter-frequency dependent separation process. Since frequency dependencies have been defined for each source, the process is able to use the frequency dependency to more accurately separate the signals. The process can use a learning algorithm that preserves frequency dependencies within each source signal, and can remove dependencies between or among the signal sources. |
| 主权项 |
1. A signal separation process, comprising:
receiving a plurality of mixed input signals at a data processing apparatus, each mixed signal being a mixture of a plurality of signal sources; using the data processing apparatus, sampling each mixed input signal using a respective rolling sampling window; using the data processing apparatus, transforming signal data in each current sampling window to frequency domain data sets; receiving the frequency domain data sets as inputs to an inter-frequency dependent separation process at the data processing apparatus; operating the inter-frequency dependent separation process at the data processing apparatus, the inter-frequency dependent separation process comprising
adapting a learning algorithm using an inter-frequency dependency; identifying, by the data processing apparatus, each component of the frequency domain data according to its correct signal source; and generating, by the data processing apparatus, a separated signal for at least one of the signal sources, wherein the inter-frequency dependent separation process uses a multivariate score function defined by the equation:φ(k)(s^i(1),…,s^i(K))=-ψ′(δλ(si))ψ(δλ(si))·qλ′(si)=ξ(δλ(si))·si(k)δλ(si)δλ(si)=(∑k(si(k)-μi(k)/σi(k))λ)1/λ wherein k represents a frequency bin within range 1 to K, φ(•) is the score function, sik is the ith source signal for frequency bin k, ŝik is the separated ith source signal for frequency k, ψ(•) is an arbitrary function, μik and σik are mean and variance, respectively, of the kth frequency bin within ith signal, q′ is first derivative of approximated probability density function q(s), δλ(s) is the λth norm of vector s, and ξ(x) is an arbitrary non-linear function of x. |
