DTMF Tone Generator: Play, Dial and Save Keypad Tones

A DTMF tone is two sine waves played together: one from a low group of four frequencies (697–941 Hz) and one from a high group of four (1209–1633 Hz), so each of the 16 telephone keys has its own pair. The generator below plays any key for as long as you hold it, dials a typed number with 100 ms tones and 100 ms gaps, and saves the sequence as a mono 16-bit WAV, all inside your browser. Those default timings fit every receiver profile listed in ITU-T Q.24, the push-button receiver recommendation.

How to use the DTMF tone generator

1. Lower the volume. Turn the system volume down before the first key. The tool starts at 5% level, about −26 dBFS peak, and stops at 25%.
2. Press a key. Hold a key on the pad to keep it sounding; a click, tap or Enter plays one tone of the set length. The readout names the row and column frequencies, and each key you press lands in the dial field. With focus inside the tool, the computer keys 0–9, *, # and A to D work too.
3. Type a sequence. Enter a number such as 555-0134 or *72#. Spaces, dashes, dots, slashes and brackets are skipped, a comma adds a 1 s pause, and the info line shows the key count, length and WAV size.
4. Pick the timing. Keep 100/100 ms, choose 70/70 for faster dialing or 200/200 for a slow sequence you can follow by ear. The Q.24 line under the fields says how many of the five published receiver profiles the setting fits, and which rule fails if it doesn't.
5. Play the sequence. Press Play sequence and watch the sounding key light up on the pad. Stop or Escape ends it at once.
6. Save the WAV. Choose 8,000 Hz for telephone work or 44,100/48,000 Hz for a music or video project, then press Save WAV. The file is built on your device at the current level, twist and timing, and Save WAV works even when the browser can't play audio.

The collapsed section at the bottom of the settings panel holds twist and the WAV sample rate. Leave twist at 0 dB unless you are checking how a decoder copes with uneven tones; the section on timing and level rules explains the window.

DTMF frequency table

The eight frequencies come from ITU-T Recommendation Q.23. Rows are the low group, columns are the high group, and each key sends the two frequencies that cross at it.

So 5 is 770 Hz plus 1336 Hz, and # is 941 Hz plus 1477 Hz. Ten of the 16 signals are digits; *, # and the fourth column are the other six. Q.23 lets a sender drift up to ±1.8% from each nominal frequency and asks for distortion products at least 20 dB below the two tones, which is why a clean sine pair decodes more reliably than a clipped one.

All eight frequencies sit well below 4 kHz, the top of the band an 8 kHz telephone channel can carry. The audio frequency chart shows where that 697–1633 Hz stretch falls against voices and instruments.

Q.24 timing and level rules

There is no single worldwide receiver spec. ITU-T Recommendation Q.24 collects the receive values five administrations adopted (NTT, AT&T, Danish, Australian and Brazilian) and treats them as examples, with national standards in charge. Read side by side, the five profiles agree more than they differ:

The tone length is the easy part: 40 ms is accepted everywhere, so any setting from 40 ms up clears it. The pause and the digit cycle decide the pace. At 100/100 ms the tool sends five keys a second and fits all five profiles with room to spare. The 70/70 preset is the fastest one here that still meets the Australian 70 ms pause and 125 ms cycle. At 50/50 the sequence fails NTT, Australian and Brazilian on the digit cycle, and Australian on its 70 ms pause too, and the Q.24 line in the card says so.

That last case matters because of a sender rule quoted in TI's application note on DSP DTMF decoding: AT&T allows at most 10 digits per second, with each tone lasting 45–55 ms inside a 100 ms slot. That's fine on an AT&T-style receiver and too fast for three of the five Q.24 profiles. I'd keep 100/100 ms for any file that will reach a system you didn't build; 70/70 is there when speed matters and you know the receiver.

Level works differently. Q.23 leaves send levels to national rules, so there is no single dBm figure to aim for, and each Q.24 receiver accepts a window per tone about 22 dB wide. The tool's level is in dBFS, relative to your device's digital full scale. What arrives on a line in dBm depends on the interface, its gain and the codec, so treat the slider as a starting point, not a line level.

The level difference between the high tone and the low tone is called twist. Lines lose more of the high group than the low group, which is why receivers accept some imbalance. AT&T allows the high tone from 4 dB above to 8 dB below the low tone; the other four profiles give a single 5, 6, 10 or 9 dB figure, and the card's Q.24 line applies that figure in both directions. At −8 dB, for example, it reports a fail for NTT and Danish.

What decoder chips accept: datasheet specs

Phone networks aren't the only receivers. Three DTMF decoder chips with public datasheets list their receive limits, and these are manufacturers' specs, not field measurements:

The figures come from the MT8870D datasheet, the Holtek HT9170 datasheet and the CM8870 datasheet. The MT8870D and CM8870 sheets state them for 40 ms tones with 40 ms pauses and an error rate better than 1 in 10,000, and the MT8870D notes that its timing depends on the external guard-time parts, so a board built around it can be slower or faster than the table.

Software decoders publish results too. TI's note lists MITEL test results for its DSP decoder: 6–7 dB of standard twist and 9 dB of reverse twist accepted, against targets above 4 and 8 dB, a 27 dB dynamic range, and 1,000 of 1,000 tone bursts detected at each of three noise levels. Put next to the chip specs, the picture is consistent: receivers are built for tones and pauses of 40 ms or more and two tones close in level. Where a spec is asymmetric, as AT&T's profile and TI's decoder are, it gives a quieter high tone more room than a louder one.

The tool's own renderer is checked with the detection method TI's note describes, a Goertzel filter for each of the eight frequencies. At 8, 44.1 and 48 kHz, the two strongest frequencies in every key are its own pair, and the other six sit more than 20 dB lower. My pick for a file that has to work on unknown hardware is 100/100 ms, 0 dB twist and a level well under clipping: it clears every Q.24 profile and every datasheet above.

Legitimate uses for DTMF tones

Four common reasons to generate DTMF tones, all on equipment you own or may test:

• Checking a phone menu or voicemail you run. Play a key sequence into your own IVR, PBX or answering system to confirm each prompt routes where it should.
• Testing a decoder board. I'd feed it a WAV at 8,000 Hz with known keys and timing, because the same file gives the same input every time, whether the receiver is an MT8870 or HT9170 module, a microcontroller project or a radio link. The MT8870D datasheet lists repeaters, mobile radio and paging among its applications.
• Sound for picture. A 48 kHz WAV drops into a video timeline at the project rate as the sound of someone dialing, with the exact keys and pace you typed.
• Learning how the pairs work. Hold one key and open the saved WAV in the audio spectrum analyzer: two peaks appear, at the row and column frequencies from the table.

Send tones only into lines and systems you own or have permission to test. The tool produces audio and nothing else, so it can't place a call or reach a system on its own.

Why a DTMF tone may not register

I'd check timing first, because it's the setting you control most directly. A tone under 40 ms, a gap under 70 ms or a digit cycle under 125 ms falls outside at least one Q.24 profile, and the Q.24 line in the card tells you which. After that, work down the signal path:

• Level. A tone too quiet for the receiver's window drops into the noise. A tone pushed into clipping creates distortion products, and TI's decoder checks second harmonics, so distortion can make a receiver reject a key it would otherwise accept.
• Twist. If the speaker, the line or an equalizer cuts the high tone hard, the pair can leave the 8–10 dB window the chips accept. Q.24 notes that line loss differs by frequency for exactly this reason.
• The codec in the path. RFC 4733 warns that low-rate voice codecs may not carry DTMF reliably, with G.723.1 as its example. The in-band fix it describes is switching to G.711, which runs at 8,000 Hz per RFC 3551.
• Events instead of audio. VoIP gateways can send keypresses as RFC 4733 events rather than as sound. If a system expects events and nothing in the path converts audio tones into events, tones played into a microphone may never register.
• Echo and noise. Q.24 notes that echo can shorten the pause a receiver hears and that noise can cause missed, split or false digits. Longer gaps help with echo; a quieter room or a direct cable helps with noise.
• A to D switched off. The MT8870D can be set to ignore the fourth column, and the HT9170's inhibit pin blocks the 1633 Hz column, so A, B, C and D may never register on a board built that way.

Holding a phone up to a laptop speaker adds both devices' frequency response and the room to the path. When a key fails that way but decodes from the WAV over a cable, the acoustic path is the problem, not the tones.

What the A, B, C and D keys are for

The fourth column, 1633 Hz, carries A, B, C and D. Q.23 defines them alongside the other 12 signals, but ordinary phones don't have them. RFC 4733 says they're less frequently encountered, typically in special applications such as military networks, and gives them event codes 12 to 15, after the digits (0 to 9), * (10) and # (11). They matter when a radio system, test set or decoder project assigns them a job, and decoder chips let a designer switch them off when it doesn't.

Limits of this DTMF tone generator

A correct pair at the right timing is necessary, not sufficient. Whether a network or device accepts the tones still depends on its national rules, its line interface, the codec and the acoustic path, and no browser tool can guarantee that.

The page doesn't dial, doesn't connect to a phone line, and never hears or records a call. It makes audio on your device and, when you ask, a WAV file. Its level is digital (dBFS), not a calibrated line level in dBm. Browsers only start audio after a press or click, as MDN's autoplay guide describes, so the first key may take a moment on a phone. A sequence holds up to 64 keys and pauses, and the comma pause is fixed at 1 s.

For one steady frequency rather than a pair, the online tone generator plays any single sine, which is handy for checking whether a speaker reproduces 697 Hz or 1633 Hz on its own.

DTMF tone generator FAQ

What frequencies does each DTMF key use?

Each key mixes one low tone (697, 770, 852 or 941 Hz, by row) with one high tone (1209, 1336, 1477 or 1633 Hz, by column). The 1 key is 697 + 1209 Hz, 0 is 941 + 1336 Hz, * is 941 + 1209 Hz and # is 941 + 1477 Hz.

How long should a DTMF tone be?

At least 40 ms, which all five ITU-T Q.24 receiver profiles accept. The pause and pace matter as much: the strictest profile asks for a 70 ms pause and a 125 ms digit cycle, so 100 ms tones with 100 ms gaps fit all five.

Does DTMF work over VoIP?

Yes, though not always as sound. VoIP systems can carry keypresses as RFC 4733 telephone events instead of audio, and low-rate codecs can garble in-band tones. In-band DTMF survives best on G.711, the 8,000 Hz telephone codec, so match the system's setting before blaming the tones.

What is twist in DTMF?

Twist is the level difference between the high-group and low-group tones of one key. AT&T's Q.24 profile accepts the high tone from 4 dB above to 8 dB below the low tone, and common decoder chips accept 8–10 dB. Equal levels, 0 dB, are the safe default.

Why do some keypads have A, B, C and D?

DTMF defines 16 signals, and the fourth column at 1633 Hz holds A, B, C and D. Standard phone keypads show only the other 12. RFC 4733 says A to D are less frequently encountered, typically in special applications such as military networks, and some decoders are set to ignore them.

Which sample rate should I save the WAV at?

8,000 Hz for phone systems and decoder tests, since the G.711 telephone codec runs at that rate and 8 kHz is enough for every DTMF frequency. Use 48,000 Hz for video and 44,100 Hz for music projects so the editor doesn't have to resample.