Speech processing

The speech processing is a scientic discipline located at the crossing of the treatment of the numeric signal and treatment of the language. There exist four big families of vocal modules:

Analyzer of word
the Voice recognition which gathers

the Voice recognition where one endeavors to include/understand what is known as and the Recognition of the speaker who allows to carry out an identification of the speaker

Voice synthesis
Codeur of word

Analogical treatment

Right from the start Telephone, the useful signal obtained by the Microphone is of a power of 1 MW (0,775 V on a impedance of 600 Ω (Ohm S) known as level 0 dB ). Knowing that it was necessary to exploit these fundamental data during decades without being able to profit from active elements (Amplification), the analog standards of transmissions were based on this reality.

This weak useful signal required an optimum quality of the cables so that the noises induced between circuits are weakest possible (plan of mixing of the pairs between segment of cable, at each point of connection in order to compensate for the imbalances of capacity measured after manufacture of the segments).

Engineers such as American Michael Pupin (1900) or Danish Emil Krarup (1902) improved the range of the signal by compensating for strong the Capacité cable by the addition of coil (uniformly distributed in the method Krarup (continuous loading) a such piano wire or induction coil every 1830 m in the method of the step Pupin (pupinization)) So by this method one increases the range of the signal, one loses on the Band-width transmitted (low-pass filter limited to 4000 Hz)

The band-width standardized for telephony is of 300 Hz with 3400 Hz . This band-width is sufficient for the intelligibility of the conversation (% of clearness is checked by measurements on random emissions of Logatome S (near to the Phonème S))

Attenuation coefficient

the telephone link undergoes the constraints of the physical parameters said primary education parameters (resistance R, inductance L, leakance G, capacity C).

the transmission analyzes on the secondary parameters which result from this:

the characteristic impedance Zc = Φ L (ω), G (ω), C (ω), ω
the exhibitor of γ transfer = α + J β

the attenuation coefficient is the term (α alpha) of the exhibitor of transfer reported to the unit of length. It is a logarithmic law comparable with the sensitivity of the human Oreille (adequacy by psophometric filters )

the dynamics of the ear is very broad:

+120 dB threshold of the pain
+90 dB level of call, cry
+60 dB level of a normal conversation
+30 dB whisper
0 dB threshold of feeling of the ear

the threshold of feeling of the ear with 1000 Hz corresponds to a variation of acoustic pressure of 20 nPa on the tympanum of the ear.

the dynamics of the levels of transmission on the phone line is as for it:

0 dB the level of the microphone of the telephone
-30 dB the level of reception on the still acceptable ear-phone

(1/1 000 of the power of the microphone)

-60 dB still perceptible noise level

(1/1 000.000 of the power of the microphone)

The sticking together between the two scales depends on the effectiveness of the microphone and the ear-phone, the distance between the ear and those (thus of the shape of compound), and of the line loss. For an entirely passive transmission (without amplification), in the best of the cases, the 0 dB of the second scale is aligned with approximately 80 dB of the first.

The first amplifications

For the first amplifications (audio AF frequency), it was necessary to solve the problems of adaptation of impedance (stabilizers on the differential transformer) in order to avoid the starting of the circuit (the effect Larsen).

First analogical multiplexed connections

For the first Multiplexing (HF high frequency), it was necessary to limit the noise levels on the white of conversation using compressor-bungee cord (noise pushed back with less than -50 dB).

Digital processing

The theorem of Shannon had determined as of 1948 qu ' it was enough to sample the basic signal with 2 times the maximum frequency of this signal to preserve the contained information in the signal to be transmitted.

It was necessary to still await a score of years to have the enough swift electronics components to exploit the first digital links of Pulse modulation Coded (MIC).

Knowing that the band-width of telephony is standardized to 300-3 400 Hz (interpreted to 0-4 000 Hz in multiplexing), it is enough to sample to 8.000 Hz to transmit the characteristics of the signal. That is to say to take a sample every 125 microseconds (1 s/8 000 Hz)

codec standardized (coder-decoders) thus work on 32 IT (time intervals) grant 3,9 microseconds to each useful way (30 telephone ways and 2 technical IT of indication) The taken sample is coded on a numerical scale of 8 bits. This scale is logarithmic curve in order to minimize the noises of quantification, coding being finer on the low levels. Two the most used codings are the µ-Law (mainly with the the United States) and a-Law (Europe). Beach from -127 to +128 (256 values is a byte)

The digital transmission is thus standardized on the MIC basic ones with 2,048 Mbits/s (256 X 8.000) and grant 64 kbits/s to each way (8 000 bytes).

These digital transmissions have the advantage of being freed from the noises of transmission. But many other problems had to be solved (coding HDB3: rapes of bipolarity and stuffing in order to preserve the rate of 2 Mbits/s, even in the absence of signals on the ways of entry).

Random links: