utbd — UTC Time Broadcast Daemon

utbd answers SIP calls — through any SIP PBX or provider, or directly — and plays a time signal in the audio format of the NIST shortwave time stations WWV and WWVH: second ticks, minute and hour markers, standard tones, the 100 Hz BCD time code and a spoken time announcement.

Service: time.tgabor.eu
Contact: info@tgabor.eu
Version: 2.0.0

This is an independent service and is not operated by or affiliated with NIST. It reproduces the broadcast format only; its station identification is its own.

---

How It Works

Telephone ─── PSTN/VoIP ─── SIP PBX / provider ─── SIP + RTP ─── utbd
                                                                   │
                                                         WWV/WWVH signal generator
                                                         (synthesised from UTC)

utbd registers with a SIP PBX or provider and answers every call routed to it, sending G.711 (A-law or µ-law) RTP. The signal is computed sample by sample from the system clock, so all callers hear the same broadcast; up to max_calls callers are served at once. The server's clock should be disciplined by NTP or GPS.

---

Broadcast Format

As described in NIST Special Publication 250-67, NIST Time and Frequency Radio Stations: WWV, WWVH, and WWVB.

Element	WWV	WWVH
Second ticks	5 ms of 1000 Hz	5 ms of 1200 Hz
Protected zone	10 ms silence before, 25 ms after each tick	same
Omitted ticks	seconds 29 and 59	same
Minute marker	800 ms of 1000 Hz	800 ms of 1200 Hz
Hour marker	800 ms of 1500 Hz	same
Tones (seconds 1–45)	500 Hz even, 600 Hz odd minutes	600 Hz even, 500 Hz odd
440 Hz	minute 2, except hour 00	minute 1, except hour 00
Silent minutes	29, 43–51, 59	0, 8–10, 14–19, 30
Station ID	minutes 0 and 30	minutes 29 and 59
Voice announcement	starts at :52.5	starts at :45
UT1 correction	doubled ticks 1–8 (+) or 9–16 (−)	same

The announcement says "At the tone, H hours, M minutes, Coordinated Universal Time." The minute marker that follows is "the tone".

Relative levels follow the station modulation: ticks and markers 100 %, voice 75 %, tones and time code 50 %.

100 Hz time code

A modified IRIG-H code, one bit per second, least significant bit first. A pulse of increased 100 Hz amplitude (3.3:1) starts 30 ms after the second: 170 ms = 0, 470 ms = 1, 770 ms = position marker. Second 0 carries no pulse; together with P0 at second 59 it marks the start of the frame. The frame refers to UTC at its start.

Seconds	Content
2	DST in effect at 00:00 UTC today
3	Leap second warning (always 0)
4–7	Year, units
9, 19, 29, 39, 49, 59	Position markers P1–P5, P0
10–13, 15–17	Minutes, units and tens
20–23, 25–26	Hours, units and tens
30–33, 35–38, 40–41	Day of year, units, tens and hundreds
50	UT1 sign (1 = add to UTC)
51–54	Year, tens
55	DST in effect at 24:00 UTC today
56–58	UT1 correction, 0.1 s units

Differences from the radio stations

• DST rules: the DST bits follow European rules by default (dst = eu). Set dst = us for the US rules used by NIST.
• Leap second warning: not set. utbd has no leap second source.
• Agency announcements: the minutes NIST uses for agency announcements (storm warnings, GPS status, geophysical alerts) carry the normal tone.
• Doubled ticks: SP 250-67 does not give the spacing of the second pulse. utbd sends it 100 ms after the first.

Over the telephone

The 100 Hz time code is below the 300–3400 Hz telephone band. It passes a G.711 VoIP path unchanged, but an analog line or handset filters it out. Ticks, markers, tones and voice are all in band.

---

Dependencies

Both are standard distribution packages:

Library	Debian / Ubuntu	Fedora / RHEL
spandsp	libspandsp-dev	spandsp-devel
Sofia-SIP	libsofia-sip-ua-dev	sofia-sip-devel

apt install build-essential pkg-config libspandsp-dev libsofia-sip-ua-dev

The Makefile also finds the Sofia-SIP headers without pkg-config. sip.c / sip.h hold the SIP user agent and RTP code, shared with actsd and actsc. For the voice files: Piper (pip install piper-tts, recommended) or espeak-ng, and optionally sox or ffmpeg for resampling.

---

Building and Installing

make                         # libraries are found with pkg-config
PIPER_MODEL=/path/to/en_US-john-medium.onnx make voice   # see "Voice files"
sudo make install-service
sudo nano /etc/utbd/utbd.conf
sudo systemctl enable --now utbd
journalctl -fu utbd

Voice files

The announcement is assembled from WAV files in voice_dir: at_the_tone, hour, hours, minute, minutes, utc, station_id, 0–19, 20, 30, 40, 50 (all .wav, 16-bit PCM, any sample rate). Without the files, utbd runs without voice announcements.

gen-voice.sh [DIR] [wwv|wwvh] creates them:

• Piper (neural TTS, natural sounding) when PIPER_MODEL points to a voice model. Recommended voices, both trained only on public-domain LibriVox recordings: en_US-john-medium (male) for WWV and en_US-kristin-medium (female) for WWVH. See License about other voices. Download the .onnx and .onnx.json files of the voice.
• espeak-ng otherwise. It is available everywhere but sounds robotic.

pip install piper-tts
PIPER_MODEL=en_US-john-medium.onnx ./gen-voice.sh voice wwv

PITCH=-3 makes the voice deeper by that many semitones (sox or ffmpeg required). Up to about −3 the words stay as clear as the original.

Recordings of a human voice saved under the same names sound best. Any sample rate works: utbd band-limits and resamples the files to 8 kHz, and trims the silence TTS engines add around each word.

---

Configuration

/etc/utbd/utbd.conf (key = value, # comments). Command-line options override the file.

Key	Default	Meaning
registrar	—	SIP PBX or provider to register with, host[:port]
username, password	—	SIP account (username required with a registrar)
allow	the registrar	Who may call in: any, or IPv4 addresses / CIDRs
sip_port	5060	Local SIP port (UDP)
rtp_ports	10200-10399	RTP port range (UDP), two ports per simultaneous call
local_ip	automatic	Local address for SIP and RTP (default: the one that routes to the registrar)
public_ip	—	Address advertised in SIP and SDP behind NAT
station	wwv	wwv or wwvh format
voice_dir	/usr/local/share/utbd/voice	Voice WAV files
dut1	0	UT1 − UTC in 0.1 s (−7..7), from IERS Bulletin A
dst	eu	DST rules for the time code: eu, us, none
advance_ms	0	Send the signal early to offset path delay
max_calls	8	Simultaneous listeners
max_call_sec	180	Call length limit

Either registrar or allow must be set.

Connecting to a PBX

utbd registers like a SIP phone; the PBX routes a number to it. Asterisk (PJSIP):

; pjsip.conf — needs a UDP transport, e.g. [transport-udp]
[utbd]
type=endpoint
context=from-internal
disallow=all
allow=alaw,ulaw
auth=utbd-auth
aors=utbd

[utbd-auth]
type=auth
auth_type=userpass
username=utbd
password=changeme

[utbd]
type=aor
max_contacts=1
qualify_frequency=60

; extensions.conf — route a number to utbd
exten => 1236,1,Dial(PJSIP/utbd)

Without a PBX, set allow to the callers' addresses and call sip:time@<utbd address> directly.

Open sip_port and rtp_ports (UDP) in the firewall for the PBX. Behind NAT, forward them and set public_ip.

Upgrading from 1.x

utbd 1.x used IAX2. Replace host/port with registrar = host[:port], peername with username and secret with password, and replace the iax.conf peer with a PJSIP endpoint as above.

---

Testing Without a PBX

--render writes the signal to an 8 kHz WAV file:

# Two minutes from the current minute
utbd --render wwv.wav --duration 120 --voice-dir voice -c /dev/null

# Around the top of an hour, WWVH format
utbd --render top.wav --start "2026-10-24 22:59" --duration 180 --station wwvh -c /dev/null

---

Security

utbd only reads its own audio files and sends audio. It does not interpret anything a caller sends: received RTP is discarded. It accepts calls only from allow (by default, the registrar's address); others get 403 Forbidden. RTP is sent only to the address negotiated in SDP. Caller ID is sanitised before logging, and calls are limited in number and length. The systemd unit runs utbd as an unprivileged user with a system-call filter and a read-only file system.

---

License

Copyright (C) 2026 Gabor Temesi info@tgabor.eu

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

The full license text is in COPYING. SPDX-License-Identifier: GPL-3.0-or-later

Third-party components

The program is linked with these libraries; no third-party source code is included in this repository.

Component	License	Project
spandsp	LGPL-2.1-only (library)	https://github.com/freeswitch/spandsp
Sofia-SIP	LGPL-2.1-or-later	https://github.com/freeswitch/sofia-sip

Both libraries may be used by GPL-3.0 programs. Distribution packages of Sofia-SIP may link OpenSSL 3 (Apache-2.0), which is compatible with GPL-3.0 as well.

Note for binary distributors. Besides its LGPL-2.1 code, the spandsp library contains a few files licensed GPL-2.0-only (image_translate.c in 0.0.6; also sprt.c and v150_1*.c in newer snapshots). GPL-2.0-only and GPL-3.0 code cannot be combined in one work. This program does not use those parts, and none of this affects building it from source or running it. If you distribute binaries, link spandsp statically (the linker then includes only the LGPL-2.1 modem objects that are used, and the LGPL-2.1 relinking terms apply) or use a spandsp build without the GPL-2.0-only files.

Voice files

gen-voice.sh runs Piper or espeak-ng as separate programs. Both are GPL-3.0-or-later; their license does not cover the speech they produce. The voice models have their own terms:

Voice	Use	Training data
en_US-john-medium	WWV (male)	Public-domain LibriVox recordings (fine-tuned from kristin)
en_US-kristin-medium	WWVH (female)	Public-domain LibriVox recordings

Avoid voices fine-tuned from lessac (such as joe, mike and ryan): the lessac dataset is licensed for research use only. The public-domain norman voice is not recommended either: it garbles most isolated words, such as the digits of the announcement.

The broadcast format follows NIST Special Publication 250-67, a work of the U.S. Government.