Training and search methods for speech recognition.

Speech recognition involves three processes: extraction of acoustic indices from the speech signal, estimation of the probability that the observed index string was caused by a hypothesized utterance segment, and determination of the recognized utterance via a search among hypothesized alternatives. This paper is not concerned with the first process. Estimation of the probability of an index string involves a model of index production by any given utterance segment (e.g., a word). Hidden Markov models (HMMs) are used for this purpose [Makhoul, J. & Schwartz, R. (1995) Proc. Natl. Acad. Sci. USA 92, 9956-9963]. Their parameters are state transition probabilities and output probability distributions associated with the transitions. The Baum algorithm that obtains the values of these parameters from speech data via their successive reestimation will be described in this paper. The recognizer wishes to find the most probable utterance that could have caused the observed acoustic index string. That probability is the product of two factors: the probability that the utterance will produce the string and the probability that the speaker will wish to produce the utterance (the language model probability). Even if the vocabulary size is moderate, it is impossible to search for the utterance exhaustively. One practical algorithm is described [Viterbi, A. J. (1967) IEEE Trans. Inf. Theory IT-13, 260-267] that, given the index string, has a high likelihood of finding the most probable utterance.

State of the art in continuous speech recognition.

In the past decade, tremendous advances in the state of the art of automatic speech recognition by machine have taken place. A reduction in the word error rate by more than a factor of 5 and an increase in recognition speeds by several orders of magnitude (brought about by a combination of faster recognition search algorithms and more powerful computers), have combined to make high-accuracy, speaker-independent, continuous speech recognition for large vocabularies possible in real time, on off-the-shelf workstations, without the aid of special hardware. These advances promise to make speech recognition technology readily available to the general public. This paper focuses on the speech recognition advances made through better speech modeling techniques, chiefly through more accurate mathematical modeling of speech sounds.

Commercial applications of speech interface technology: an industry at the threshold.

Speech interface technology, which includes automatic speech recognition, synthetic speech, and natural language processing, is beginning to have a significant impact on business and personal computer use. Today, powerful and inexpensive microprocessors and improved algorithms are driving commercial applications in computer command, consumer, data entry, speech-to-text, telephone, and voice verification. Robust speaker-independent recognition systems for command and navigation in personal computers are now available; telephone-based transaction and database inquiry systems using both speech synthesis and recognition are coming into use. Large-vocabulary speech interface systems for document creation and read-aloud proofing are expanding beyond niche markets. Today's applications represent a small preview of a rich future for speech interface technology that will eventually replace keyboards with microphones and loud-speakers to give easy accessibility to increasingly intelligent machines.

Voice-processing technologies--their application in telecommunications.

As the telecommunications industry evolves over the next decade to provide the products and services that people will desire, several key technologies will become commonplace. Two of these, automatic speech recognition and text-to-speech synthesis, will provide users with more freedom on when, where, and how they access information. While these technologies are currently in their infancy, their capabilities are rapidly increasing and their deployment in today's telephone network is expanding. The economic impact of just one application, the automation of operator services, is well over $100 million per year. Yet there still are many technical challenges that must be resolved before these technologies can be deployed ubiquitously in products and services throughout the worldwide telephone network. These challenges include: (i) High level of accuracy. The technology must be perceived by the user as highly accurate, robust, and reliable. (ii) Easy to use. Speech is only one of several possible input/output modalities for conveying information between a human and a machine, much like a computer terminal or Touch-Tone pad on a telephone. It is not the final product. Therefore, speech technologies must be hidden from the user. That is, the burden of using the technology must be on the technology itself. (iii) Quick prototyping and development of new products and services. The technology must support the creation of new products and services based on speech in an efficient and timely fashion. In this paper I present a vision of the voice-processing industry with a focus on the areas with the broadest base of user penetration: speech recognition, text-to-speech synthesis, natural language processing, and speaker recognition technologies. The current and future applications of these technologies in the telecommunications industry will be examined in terms of their strengths, limitations, and the degree to which user needs have been or have yet to be met. Although noteworthy gains have been made in areas with potentially small user bases and in the more mature speech-coding technologies, these subjects are outside the scope of this paper.

A zinc finger directory for high-affinity DNA recognition

We have used two monovalent phage display libraries containing variants of the Zif268 DNA-binding domain to obtain families of zinc fingers that bind to alterations in the last 4 bp of the DNA sequence of the Zif268 consensus operator, GCG TGGGCG. Affinity selection was performed by altering the Zif268 operator three base pairs at a time, and simultaneously selecting for sets of 16 related DNA sequences. In this way, only four experiments were required to select for all possible 64 combinations of DNA triplet sequences. The results show that (i) for high-affinity DNA binding in the range observed for the Zif268 wild-type complex (Kd = 0.5–5 nM), finger 1 specifically requires the arginine at the carboxy terminus of its recognition helix that forms a bidentate hydrogen-bond with the guanine base (G) in the crystal structure of Zif268 complexed to its DNA operator sequence GCG TGG GCG; (ii) when the guanine base (G) is replaced by A, C, or T, a lower-affinity family (Kd ⩾ 50 nM) can be detected that shows an overall tendency to bind G-rich DNA; (iii) the residues at position 2 on the finger 2 recognition helix do not appear to interact strongly with the complementary 5′ base in the finger 1 binding site; and (iv) unexpected substitutions at the amino terminus of finger 1 can occasionally result in specificity for the 3′ base in the finger 1 binding site. A DNA recognition directory was constructed for high-affinity zinc fingers that recognize all three bases in a DNA triplet for seven sequences of the type GNN. Similar approaches may be applied to other zinc fingers to broaden the scope of the directory.

Lune/eye gone, a Pax-like protein, uses a partial paired domain and a homeodomain for DNA recognition

Pax proteins, characterized by the presence of a paired domain, play key regulatory roles during development. The paired domain is a bipartite DNA-binding domain that contains two helix–turn–helix domains joined by a linker region. Each of the subdomains, the PAI and RED domains, has been shown to be a distinct DNA-binding domain. The PAI domain is the most critical, but in specific circumstances, the RED domain is involved in DNA recognition. We describe a Pax protein, originally called Lune, that is the product of the Drosophila eye gone gene (eyg). It is unique among Pax proteins, because it contains only the RED domain. eyg seems to play a role both in the organogenesis of the salivary gland during embryogenesis and in the development of the eye. A high-affinity binding site for the Eyg RED domain was identified by using systematic evolution of ligands by exponential enrichment techniques. This binding site is related to a binding site previously identified for the RED domain of the Pax-6 5a isoform. Eyg also contains another DNA-binding domain, a Prd-class homeodomain (HD), whose palindromic binding site is similar to other Prd-class HDs. The ability of Pax proteins to use the PAI, RED, and HD, or combinations thereof, may be one mechanism that allows them to be used at different stages of development to regulate various developmental processes through the activation of specific target genes.

Effects of muscarinic blockade in perirhinal cortex during visual recognition

Stimulus recognition in monkeys is severely impaired by destruction or dysfunction of the perirhinal cortex and also by systemic administration of the cholinergic-muscarinic receptor blocker, scopolamine. These two effects are shown here to be linked: Stimulus recognition was found to be significantly impaired after bilateral microinjection of scopolamine directly into the perirhinal cortex, but not after equivalent injections into the laterally adjacent visual area TE or into the dentate gyrus of the overlying hippocampal formation. The results suggest that the formation of stimulus memories depends critically on cholinergic-muscarinic activation of the perirhinal area, providing a new clue to how stimulus representations are stored.

Monoclonal antibodies against oxidized low-density lipoprotein bind to apoptotic cells and inhibit their phagocytosis by elicited macrophages: Evidence that oxidation-specific epitopes mediate macrophage recognition

Apoptosis is recognized as important for normal cellular homeostasis in multicellular organisms. Although there have been great advances in our knowledge of the molecular events regulating apoptosis, much less is known about the receptors on phagocytes responsible for apoptotic cell recognition and phagocytosis or the ligands on apoptotic cells mediating such recognition. The observations that apoptotic cells are under increased oxidative stress and that oxidized low-density lipoprotein (OxLDL) competes with apoptotic cells for macrophage binding suggested the hypothesis that both OxLDL and apoptotic cells share oxidatively modified moieties on their surfaces that serve as ligands for macrophage recognition. To test this hypothesis, we used murine monoclonal autoantibodies that bind to oxidation-specific epitopes on OxLDL. In particular, antibodies EO6 and EO3 recognize oxidized phospholipids, including 1-palmitoyl 2-(5-oxovaleroyl) phosphatidylcholine (POVPC), and antibodies EO12 and EO14 recognize malondialdehyde-lysine, as in malondialdehyde-LDL. Using FACS analysis, we demonstrated that each of these EO antibodies bound to apoptotic cells but not to normal cells, whereas control IgM antibodies did not. Confocal microscopy demonstrated cell-surface expression of the oxidation-specific epitopes on apoptotic cells. Furthermore, each of these antibodies inhibited the phagocytosis of apoptotic cells by elicited peritoneal macrophages, as did OxLDL. In addition, an adduct of POVPC with BSA also effectively prevented phagocytosis. These data demonstrate that apoptotic cells express oxidation-specific epitopes—including oxidized phospholipids—on their cell surface, and that these serve as ligands for recognition and phagocytosis by elicited macrophages.

Tn5/IS50 target recognition

This communication reports an analysis of Tn5/IS50 target site selection by using an extensive collection of Tn5 and IS50 insertions in two relatively small regions of DNA (less than 1 kb each). For both regions data were collected resulting from in vitro and in vivo transposition events. Since the data sets are consistent and transposase was the only protein present in vitro, this demonstrates that target selection is a property of only transposase. There appear to be two factors governing target selection. A target consensus sequence, which presumably reflects the target selection of individual pairs of Tn5/IS50 bound transposase protomers, was deduced by analyzing all insertion sites. The consensus Tn5/IS50 target site is A-GNTYWRANC-T. However, we observed that independent insertion sites tend to form groups of closely located insertions (clusters), and insertions very often were spaced in a 5-bp periodic fashion. This suggests that Tn5/IS50 target selection is facilitated by more than two transposase protomers binding to the DNA, and, thus, for a site to be a good target, the overlapping neighboring DNA should be a good target, too. Synthetic target sequences were designed and used to test and confirm this model.

Suppressor mutations in Escherichia coli methionyl-tRNA formyltransferase: Role of a 16-amino acid insertion module in initiator tRNA recognition

The specific formylation of initiator methionyl-tRNA by methionyl-tRNA formyltransferase (MTF; EC 2.1.2.9) is important for the initiation of protein synthesis in eubacteria and in eukaryotic organelles. The determinants for formylation in the tRNA are clustered mostly in the acceptor stem. As part of studies on the molecular mechanism of recognition of the initiator tRNA by MTF, we report here on the isolation and characterization of suppressor mutations in Escherichia coli MTF, which compensate for the formylation defect of a mutant initiator tRNA, lacking a critical determinant in the acceptor stem. We show that the suppressor mutant in MTF has a glycine-41 to arginine change within a 16-amino acid insertion found in MTF from many sources. A mutant with glycine-41 changed to lysine also acts as a suppressor, whereas mutants with changes to aspartic acid, glutamine, and leucine do not. The kinetic parameters of the purified wild-type and mutant Arg-41 and Lys-41 enzymes, determined by using the wild-type and mutant tRNAs as substrates, show that the Arg-41 and Lys-41 mutant enzymes compensate specifically for the strong negative effect of the acceptor stem mutation on formylation. These and other considerations suggest that the 16-amino acid insertion in MTF plays an important role in the specific recognition of the determinants for formylation in the acceptor stem of the initiator tRNA.

Toward rules for 1:1 polyamide:DNA recognition

Polyamides composed of four amino acids, imidazole (Im), pyrrole (Py), hydroxypyrrole (Hp), and β-alanine (β), are synthetic ligands that form highly stable complexes in the minor groove of DNA. Although specific pairing rules within the 2:1 motif can be used to distinguish the four Watson⋅Crick base pairs, a comparable recognition code for 1:1 polyamide:DNA complexes had not been described. To set a quantitative baseline for the field, the sequence specificities of Im, Py, Hp, and β for the four Watson⋅Crick base pairs were determined for two polyamides, Im-β-ImPy-β-Im-β-ImPy-β-Dp (1, for Im, Py, and β) and Im-β-ImHp-β-Im-β-ImPy-β-Dp (2, for Hp), in a 1:1 complex within the DNA sequence context 5′-AAAGAGAAGAG-3′. Im residues do not distinguish G,C from A,T but bind all four base pairs with high affinity. Py and β residues exhibit ≥10-fold preference for A,T over G,C base pairs. The Hp residue displays a unique preference for a single A⋅T base pair with an energetic penalty.

Synaptic pattern formation during cellular recognition

Cell–cell recognition often requires the formation of a highly organized pattern of receptor proteins (a synapse) in the intercellular junction. Recent experiments [e.g., Monks, C. R. F., Freiberg, B. A., Kupfer, H., Sciaky, N. & Kupfer, A. (1998) Nature (London) 395, 82–86; Grakoui, A., Bromley, S. K., Sumen, C., Davis, M. M., Shaw, A. S., Allen, P. M. & Dustin, M. L. (1999) Science 285, 221–227; and Davis, D. M., Chiu, I., Fassett, M., Cohen, G. B., Mandelboim, O. & Strominger, J. L. (1999) Proc. Natl. Acad. Sci. USA 96, 15062–15067] vividly demonstrate a complex evolution of cell shape and spatial receptor–ligand patterns (several microns in size) in the intercellular junction during immunological synapse formation. The current view is that this dynamic rearrangement of proteins into organized supramolecular activation clusters is driven primarily by active cytoskeletal processes [e.g., Dustin, M. L. & Cooper, J. A. (2000) Nat. Immunol. 1, 23–29; and Wulfing, C. & Davis, M. M. (1998) Science 282, 2266–2269]. Here, aided by a quantitative analysis of the relevant physico-chemical processes, we demonstrate that the essential characteristics of synaptic patterns observed in living cells can result from spontaneous self-assembly processes. Active cellular interventions are superimposed on these self-organizing tendencies and may also serve to regulate the spontaneous processes. We find that the protein binding/dissociation characteristics, protein mobilities, and membrane constraints measured in the cellular environment are delicately balanced such that the length and time scales of spontaneously evolving patterns are in near-quantitative agreement with observations for synapse formation between T cells and supported membranes [Grakoui, A., Bromley, S. K., Sumen, C., Davis, M. M., Shaw, A. S., Allen, P. M. & Dustin, M. L. (1999) Science 285, 221–227]. The model we present provides a common way of analyzing immunological synapse formation in disparate systems (e.g., T cell/antigen-presenting cell junctions with different MHC-peptides, natural killer cells, etc.).

Genetic evidence that the RAG1 protein directly participates in V(D)J recombination through substrate recognition.

RAG1 protein is essential for the activation of V(D)J recombination in developing lymphocytes (V, variable; D, diversity; J, joining). However, it has not been determined whether its role involves substrate recognition and catalysis. A single amino acid substitution mutation in the RAG1 gene has now been identified that renders its activity sensitive to the sequence of the coding region abutting the heptamer site in the recombination signal sequence. These results strongly imply that RAG1 interacts directly with DNA.