SU403090A1 - FOLLOWING MAIN TONE OF SPEECH - Google Patents

Description

The invention relates to the field of measuring electrical signals and can be used in speech compression systems, in automatic speech recognition systems, as well as in phonetic studies of speech intonation.

The following speech pitch selectors are known, consisting of a preselector, a compressor, a tunable filter, a limiter, a single vibrator, and a low pass filter. At the output of the low-pass filter, a voltage proportional to the frequency of the fundamental tone is generated. This voltage is used to control the tuning frequency of the bandpass filter. As a result, the bandpass filter monitors the pitch frequency.

The prior art devices have a low reliability of distinguishing the pitch of speech, since they are prone to excitation due to the fact that feedback is carried out in the same frequency range as the selection of the frequency of the pitch. In addition, changes in the characteristics of the feedback due to various destabilizing factors lead to inaccuracies in tuning the tunable filter, and these errors are not compensated by the feedback.

After pauses, there is a significant detuning of the tunable filter relative to the pitch frequency due to the small memory determined by the time of the low pass filter transition process in the control circuit. Increasing the memory by narrowing the bandwidth leads to poor tracking of fast pitch changes.

The aim of the invention is to improve the reliability of the pitch. This is achieved by the addition of a serially connected modulator, a first bandpass filter, a second bandpass filter, a demodulator, and a low pass filter between the compressor and the limiter. Output

through the amplifier, the first bandpass filter is connected to the first input of the phase detector, to the second of which, through the limiter, the output of the second bandpass filter is connected, and the output of the phase detector through

parallel connectors integrator and low-pass filter connected to the second input of the second band-pass filter and to the input of the generator, the output of which is connected to the input of the modulator and to the input of the demodulator.

The drawing shows a block diagram of the trace 11. His pitch extractor.

device contains preselector 1, compressor 2, modulator 3, generator 4, first bandpass filter 5, second bandpass filter

6, amplifier 7, limiter 8, phase detector 9, low-pass filter 10, integrator 11, demodulator 12, low-pass filter 13, stop 14, one-shot 15, low-pass filter 16. The speech signal is fed to preselector 1 (a nanoscale and bandpass filter with cutoff frequencies Hz and fu 740 Hz), which allocates a frequency band corresponding to the base tone. The signal obtained at the output of the preselector is fed to compressor 2, normalized to the level and fed to modulator 3, to the other input of which a signal is received from generator 4, with the result that the signal is transferred to the high-frequency region and two sidebands are formed near the generator / gay frequency the spectrum of the speech signal transmitted through the preselector and compressor. The oscillation frequency of the generator should be: / gene (for example, /, Hz, and 2 / in 2-740: 1480 Hz). From the modulator, the signal arrives at the first band-pass filter 5 with the transmission coefficient / Ci (/), which suppresses the generator frequency and the other sideband, and in the selectable sideband provides the preferential passage to the frequency corresponding to the lowest frequency harmonic of the fundamental tone (FRT frequency), when attenuating frequencies corresponding to higher frequency harmonics. From the output of the first bandpass filter, the signal is fed to a second bandpass filter 6 with a Kzil transfer coefficient) tuned to the frequency: / cut.pfg - / gene / n. Signals From the output of the first bandpass filter, through the amplifier 7 and from the output of the second bandpass filter, through the limiter 8 phase detector 9, where their phases are compared. It is known that the spectrum of the tonal portions of speech is a set of discretes at frequencies corresponding to the harmonics of the fundamental tone. At the output of the first bandpass filter, the spectrum will repeat relative to the generator frequency fren the spectrum of the original speech signal passed through the preselector and compressor if the filter covers the upper sideband and is reverse if the filter covers the lower sideband. In this example, the spectrum at the filter output is inverse, since the filter covers the lower band. If the frequency of the fundamental tone is equal to the lower cutoff frequency of the preselector: / from - / n, then the signal on the output of the second bandpass filter corresponds to its resonant frequency, therefore the phase shift does not occur and the voltage at the output of the phase detector is equal to a bullet. The output signal of the phase detector through a low-pass filter (10 Hz) and integrator 11 controls the generator with a large constant time (s). Since the voltage at the output of the phase detector is zero, the generator frequency remains unchanged. The signal from the output of the second bandpass filter is supplied to the demodulator 12, to the other input of which a signal is supplied from the generator. At the output of the demodulator, a signal is formed, the spectrum of which consists of discretes at the frequency fi and / 2: 11 - (gene G Dream. Pfg / 2 - gene /; / 1 - / gene / o / 2; - / from Low Frequency / 2 is the frequency of the fundamental tone of speech. With the help of a low-pass filter 13, the frequency of the pitch tone is separated out / from the output of the low-pass filter is limited by limiter 14 and a single-pulse 15 is output. Thus, a single-pulse sequence is formed at the output of the single-vibrator corresponding to frequencies At the filter output (/ cf 50 Hz), a voltage proportional to the frequency of the fundamental tone is created at the filter output (/ cf. 50 Hz). If the frequency of the RT is changed, the frequency of the signal at the output of the second frequency filter will be different from its resonant frequency Therefore, the signal at the output of the second bandpass filter acquires a phase shift determined by the phase characteristic of this filter.This shift is detected and converted by the phase detector into a proportional voltage, which is often and integrator adjusts the oscillator frequency is approached towards the oscillation signal at the output of the second bandpass filter the resonance frequency of this filter. In this way, the frequency of the fundamental tone is monitored, and feedback compensates for changes in the characteristics of its individual elements, which occur through iodine by destabilizing factors. The higher the pitch frequency, the greater the voltage at the output of the phase detector, the wider the passband of the second bandpass filter, and the lower the pitch frequency, the narrower the passband of the second bandpass filter. The integrator maintains the tracking system for the duration of the pauses in a position close to that which followed the tracking of the fundamental tone in the previous tone section.

Subject invention

A tracking pitch extractor comprising a series-connected preselector, compressor, limiter, one-shot and low-pass filter, characterized in that, in order to increase the reliability of the pitch, a serially connected modulator, first band-pass filter, second band-pass filter, demodulator, low pass filter.

the output of the first band-pass filter is connected via an amplifier to the first input of the phase detector, to the second input of which the output of the second band-pass filter is connected through a limiter, and the output of the phase detector is connected via a parallel-connected integrator and low pass filter to the second input of the second band filter and to the generator input, the output of which is connected to the input of the modulator and to the input of the demodulator.