FreeDV digital voice
Open-source digital voice for HF hams, including the newer RADE mode built with machine learning.
What does it sound like?
A steady rushing hiss about as wide as a voice signal, with no speech you can make out on an ordinary radio.
Who uses it?
Amateur radio operators who want clear, noise-free voice on the shortwave ham bands.
Where and when can I hear it?
On the HF ham bands, mostly 20 meters. The FreeDV project's live station list, FreeDV Reporter (qso.freedv.org), shows who is on the air and on which frequencies, and lists the current calling frequencies. Busiest on weekends and during FreeDV Activity Days (the third weekend of each month).
Hear it live
- University of Twente WebSDR (Enschede, the Netherlands)
No fixed frequency. FreeDV Reporter lists stations on the air right now with their frequencies; enter one in the receiver in USB. The busiest time is FreeDV Activity Day, the third weekend of each month.
These are free public receivers run by other people; we only link to them. You'll hear the signal only if it's on the air right now and conditions carry it to that receiver, so if it's quiet, try another time. Links last reviewed 2026-09-28.
The science
FreeDV turns speech into a digital signal a normal SSB radio can carry. Its older modes pair the open Codec 2 speech coder (700 to 3,200 bits per second) with a modem and error correction. The newer RADE mode (Radio Autoencoder) combines machine learning with classic DSP: a neural network turns speech features (spectrum, pitch, voicing) into continuously valued QAM symbols sent on many carriers, and the receiver rebuilds the voice with the FARGAN neural vocoder; it occupies about 1.5 kHz but carries about 8 kHz of audio bandwidth.
Why use it?
RADE is reported to deliver good speech at signal-to-noise ratios as low as -2 dB while using about half the spectrum of a typical 3 kHz SSB signal. The software is open source throughout, with no proprietary codec.
Lore
Codec 2 traces back to its author David Rowe's 1997 PhD thesis on harmonic sinusoidal speech coding. RADE, developed with grant funding from Amateur Radio Digital Communications, was released in June 2025, according to ARDC.
Easily confused with
- Single-sideband voice (SSB): plain SSB voice is understandable speech
- MT63: MT63 is text, not voice