中科微点-全球专利检索

  1. Login
  2. Sign Up

Search Results

122 records (took 0.02 seconds)

    Speech analyzing system
    20.
    发明授权
    Speech analyzing system 失效
    语音分析系统

    公开(公告)号：US3296374A

    公开(公告)日：1967-01-03

    申请号：US29134463

    申请日：1963-06-28

    Applicant: IBM

    Inventor： CLAPPER GENUNG L

    IPC: G10L19/00 ,  G10L19/02

    CPC classification number: G10L19/00 ,  G10L19/02

    Abstract: 1,020,527. Automatic speech recognition. INTERNATIONAL BUSINESS MACHINES CORPORATION. May 29, 1964 [June 28, 1963], No. 22279/64. Heading G4R. In sound analysing apparatus the formant frequencies are found by applying the sound signal to a plurality of filters each responsive to one of a number of frequency bands, there being means for indicating which filter output has the maximum value compared with the outputs of the filters responsive to adjacent frequency bands. The signal from microphone 1, Fig. 1, is compressed in pre-amplifier 2 controlled by A.G.C. unit 35. The compressed signal passes to a bank of filters F1-F15. The top filter F1 responds to " noise," i.e. fricative or sibilant signals, and consists of a resonant L-C circuit. The other filters respond to narrow bands, the centre frequency of which is indicated for each. These circuits each comprise a twin-T filter network and a difference amplifier. The output of filter F1 is applied via a coupling capacitor 69 and inverter amplifier 70 to an integrating pulseshaper 130 which recognizes the presence of fricative sound frequencies. The outputs of filters F2-F15 are coupled to rectifiers R1- R14 and the outputs, representing the energy in that particular frequency band, are applied to balance detectors B1-B13. An output is provided from each balance detector on one of two leads depending on the relationship between the energies in the corresponding two adjacent channels. Gates 120 connected to adjacent balance detectors give an output to indicate a local maximum in the curve of energy distribution over the frequency bands. These maxima, which indicate the formants of the speech wave, are applied via integrating pulse shapers 130. In units 180 transients are detected and combined at 205 to obtain timing pulses for sampling the outputs in line drivers 200. The signals are entered in a matrix store 270 as in Specification 966,211. If the A.G.C. signal exceeds a certain value a lamp 36 lights. The signal is also applied to a slope detector 145, a positive or negative change exceeding a certain value passing a signal via unit 180 for entry into the matrix store.

Patent Agency Ranking