This paper brie y describes the systems presented by the Software Technologies Working Group (http://gtts.ehu.es, GTTS) of the University of the Basque Country (UPV/EHU) to the Query-by-Example Search on Speech Task (QUESST) at MediaEval 2014. The GTTS-EHU systems consist of four modules: (1) feature extraction; (2) speech activity detection; (3) DTW-based query matching; and (4) score calibration and fusion. The submitted systems follow the same approach used in our SWS 2013 submissions, with two minor changes (needed to address the new task): the search stops at the most likely query detection (no further detections are looked for) and a score is produced for each (query, document) pair. The two approximate matching types introduced in QUESST have not received special treatment. This year, we have just explored the use of reduced feature sets, obtaining worse results but at lower computational costs.
The MediaEval 2014 Query-by-Example Search on Speech
Task (QUESST) consists of searching for a spoken query
within a set of spoken documents. For each pair (query,
document), a score in the range ( 1; +1) must be produced,
the higher (the more positive) the score, the more likely that
the query appears in the document. System performance is
primarily measured in terms of a normalized cross-entropy
cost Cnxe. Term-Weighted Value metrics (ATWV/MTWV)
are used as secondary metrics, along with the processing
resources (real-time factor and peak memory usage) required
by the submitted systems. For more details on QUESST,
see [
The Brno University of Technology (BUT) phone decoders
for Czech, Hungarian and Russian [
Given an input signal of length T , the decoder outputs the posterior probability of each state s (1 s S) of each unit i (1 i M ) at each frame t (1 t T ), pi;s(t), where M is the number of units and S the number of states per unit. The posterior probability of each unit i at each frame t are computed by adding the posteriors of its states: pi(t) =
X pi;s(t) 8s (1) Finally, the posteriors of the three non-phonetic units are added and stored as a single non-speech posterior. Thus, the size of the frame-level feature vectors is 43, 59 and 50 for the Czech, Hungarian and Russian BUT decoders, respectively. 2.1.1
Reduced feature sets
In [
Given an audio signal, Speech Activity Detection (SAD) is performed by discarding those phone posterior feature vectors for which the non-speech posterior is the highest. The remaining vectors, along with their corresponding time o sets, are stored for further use, but the component corresponding to the non-speech unit is deleted. If the number of speech vectors is too low (in this evaluation, 10, meaning 0.1 seconds), the whole signal is discarded (thus saving time and possibly avoiding many false alarms) and a oor score is output (in this evaluation, 10 5). 2.3
Given two SAD- ltered sequences of feature vectors corresponding to a spoken query q and a spoken document x, the cosine distance is computed between each pair of vectors, q[i] and x[j] as follows: d(q[i]; x[j]) = log Note that d(v; w) 0, with d(v; w) = 0 if and only if v and w are perfectly aligned and d(v; w) = +1 if and only if v and w are orthogonal. The distance matrix computed according to Eq. 2 is further normalized with regard to the spoken document x, as follows: d(q[i]; x[j]) dmin(i) dmax(i) dmin(i) with dmin(i) = min d(q[i]; x[j]) and dmax(i) = max d(q[i]; x[j]). j j (2) (3) In this way, matrix values are all comprised between 0 and 1, so that a perfect match would produce a quasi-diagonal sequence of zeroes.
The best match of a query q of length m in a spoken document x of length n is de ned as that minimizing the average distance in a crossing path of the matrix dnorm. A crossing path starts at any given frame of x, k1 2 [1; n], then traverses a region of x which is optimally aligned to q (involving L vector alignments), and ends at frame k2 2 [k1; n]. The average distance in this crossing path is:
L davg(q; x) = 1 X dnorm(q[il]; x[jl]) (4)
L where il and jl are the indices of the vectors of q and x in the alignment l, for l = 1; 2; : : : ; L. Note that i1 = 1, iL = m, j1 = k1 and jL = k2. The minimization operation is accomplished by means of a dynamic programming procedure, which is (n m d) in time (d: size of feature vectors) and (n m) in space. The detection score is computed as 1 davg(q; x). Once the best match is obtained, the search procedure stops. As noted above, if either q or x have not enough speech samples, no alignment is performed and a oor score (10 5) is output. Note that a detection score must be mandatorily produced for each pair (q; x).
First, the so-called q-norm (query normalization) is
applied, so that zero-mean and unit-variance scores are
obtained per query [
Table 1 shows the performance and processing costs of GTTS-EHU systems on QUESST 2014. To speed up computations, experiments with the full and reduced sets of features were carried out on di erent machines (see Table 1), which makes it nonsense to compare the reported times.