Talk off in DTMF telephone sets

Talk-off

Talk-off is the false detection of a DTMF digit by the exchange's DTMF receiver when speech, music or noise on the line happens to contain energy at a low-group and a high-group DTMF frequency at the same time with enough level and duration. The receiver then wrongly "hears" a digit, which can disturb a call in progress or a call being set up (e.g. IVR menus). Its opposite, talk-down, is when speech or noise stops a real tone from being detected.

Design features to protect against talk-off

1. Choice of frequencies

• Eight tones in two groups: low 697, 770, 852, 941 Hz; high 1209, 1336, 1477, 1633 Hz.
• Each digit uses exactly one tone from each group (2-out-of-8 code). Speech rarely produces two pure, steady tones from both groups simultaneously.
• Frequencies are not harmonically related: no tone is a multiple (2nd, 3rd harmonic) of another, so harmonics of voice or of one tone cannot create a valid pair.
• Intermodulation products (sum and difference frequencies) of any two tones do not fall on other DTMF frequencies.
• Spacing of about 10% between adjacent tones, matching filter selectivity.
• High-group tones lie above the strongest speech energy (speech formant energy is mostly below 1 kHz).

2. Receiver checks (guard features)

• Band-split filters: low and high groups are separated, limited and detected separately; the receiver checks that one and only one tone appears in each group.
• Minimum tone duration (about 40 ms) and minimum inter-digit pause (about 40 ms): short bursts in speech are ignored.
• Frequency tolerance: accept within ±1.5%, reject beyond ±3.5%.
• Signal-to-total-energy (guard) check: the two tones must contain most of the energy in the band; if much energy is present at other frequencies (as in speech), detection is blocked.
• Twist limits: the level difference between high and low tones must be within limits (e.g. high tone up to ~4 dB above or ~8 dB below the low tone).
• Minimum level threshold (e.g. −25 dBm) so weak background signals are ignored.
• Third harmonic/second harmonic checks in some receivers to reject speech.

3. Telephone set features

• Crystal-controlled oscillator gives accurate frequencies and steady amplitude.
• Microphone muted while a key is pressed, so voice is not mixed with tones.
• Proper sending level (about −6 to −8 dBm per tone) and pre-emphasis of the high group by ~2 dB to make up for line loss.
• Standard timing: tone ≥ 40–50 ms, pause ≥ 40–50 ms.

4. Network/exchange features

• The DTMF receiver is normally connected only during the dialling phase and disconnected after call set-up, so conversation cannot cause false digits (except for services that need end-to-end tones).

With these measures, the probability of talk-off becomes very small (standards such as Bellcore require only a few false detections over many hours of recorded speech) while genuine digits are detected reliably.