Every digital voice mode faces the same problem the instant you key up: your voice, as raw audio, is far too fat to fit down a narrow radio channel. Something has to shrink it — drastically — and rebuild it on the far end. That something is a vocoder, and almost the entire hobby runs on one of two families: AMBE and Codec2. Which family a mode uses explains a surprising amount, right down to why some setups need a little USB stick with a chip inside. (New to the modes themselves? Start with the digital modes guide.)
Ordinary telephone-quality audio runs about 64 kilobits per second. A digital voice RF channel, after it spends bits on error correction and framing, leaves you only a few kilobits per second for the voice itself. That's better than a tenfold reduction, and you can't get there by “compressing the sound” the way a ZIP file compresses a document. A vocoder gets there by a cleverer trick: instead of describing the waveform, it describes the speaker. Many times a second it analyzes the shape of the vocal tract producing the sound — the pitch, the resonances, the voiced-versus-unvoiced texture — and transmits those parameters. The far end runs the model in reverse and synthesizes fresh speech that sounds like you. That's why digital voice has its characteristic texture: you aren't hearing a recording of your voice, you're hearing a reconstruction built from its blueprint.
AMBE — Advanced Multi-Band Excitation — is the product of Digital Voice Systems, Inc. (DVSI), and its improved version, AMBE+2, sits under most of the hobby's modes. D-STAR uses the earlier AMBE generation (its so-called 2020 vocoder); DMR, System Fusion (C4FM), and NXDN use AMBE+2. P25 is the exception, and it needs a section of its own — see below. It is genuinely good — intelligible speech at remarkably low bitrates — and it is genuinely proprietary. DVSI's implementation is licensed technology, and the reference version ships as a physical chip. Open-source reimplementations of the decoder exist, but licensing and intellectual-property concerns mean that for full-quality, above-board encoding and decoding, the standard answer for years has been a hardware dongle with a DVSI chip inside: the ThumbDV, the DVStick 30, the DVStick 33, and their relatives. Hold that thought — it's the entire reason the transcoding page exists.
There is a third family, and leaving it out is the most common mistake made about digital voice.
IMBE — Improved Multi-Band Excitation — is the vocoder of P25 Phase 1, which is what essentially all amateur P25 runs. It is an older member of the same DVSI multi-band-excitation line as AMBE, and like AMBE it was proprietary in origin.
But its patent has expired.
That single fact changes the economics completely. An open-source IMBE vocoder exists and has for years — the OP25 project has shipped one under the GPL since 2008 — and the modern URF transcoder, tcd, uses it. P25 transcodes in software, on an ordinary CPU, with no dongle and no licence.
Note carefully what expiring does and does not do. IMBE is still proprietary — DVSI designed it and it is their specification. What lapsed is the patent, and with it the legal obstacle to writing your own implementation. Proprietary and patented are not the same word, and the difference is worth exactly one dongle.
One fact explains most of the arguments about which mode uses which codec, and it is almost never stated plainly.
AMBE+2 at full rate is backwards-compatible with IMBE. Both occupy the same 7,200 bps slot; only the encoder changed.
So when one source says P25 is IMBE and another says AMBE+2, they are not disagreeing. They are describing the same bits from opposite ends.
Codec2 is the work of David Rowe, VK5DGR, and it is everything AMBE is not in terms of access: free, open-source, royalty-free, and running entirely in software on any modest processor. It comes as a family of bitrates — 3200 bits per second down through 1600, 700, and lower — trading fidelity for narrowness as you descend. In amateur digital voice it is the voice of M17, which uses Codec2 at 3200 bps. Because there is no chip to buy and no license to sign, anyone can build a mode on it, read the source, and run the encoder on a Raspberry Pi with cycles left over. That openness is a founding principle of M17 specifically: an all-open digital voice mode, vocoder included.
| AMBE / AMBE+2 | IMBE | Codec2 | |
|---|---|---|---|
| Origin | DVSI (commercial) | DVSI (commercial) | David Rowe, VK5DGR (open source) |
| License | Proprietary, patented / licensed | Proprietary — patent EXPIRED | Open, royalty-free (LGPL) |
| Runs in | Hardware chip (dongle) is the standard | Software — free implementations exist | Pure software |
| Used by | D-STAR, DMR, YSF / C4FM, NXDN | P25 (Phase 1) | M17 |
| Typical rate | ~2.4–3.6 kbit/s including FEC | 7200 bps (4400 voice + 2800 FEC) | 3200 / 1600 / 700 bps family |
| Cost to a homebrewer | Buy a dongle | Free | Free |
The practical upshot lands squarely on anyone building infrastructure. Because the AMBE-based modes dominate the hobby, and because their codec lives behind a licensing wall, an enormous amount of digital-voice plumbing revolves around one question: how do you get audio into and out of AMBE? The moment two modes that don't share a vocoder need to talk to each other, something in the middle has to decode one and re-encode the other — and if either side is AMBE, that means a chip in the loop. That job is transcoding, and it's the natural next stop.
The big idea
A vocoder doesn't compress your voice; it models your voice and sends the blueprint. Three families divide amateur digital voice: AMBE, proprietary and chip-bound, under D-STAR, DMR, C4FM and NXDN; IMBE, proprietary in origin but with an expired patent, which is why P25 transcodes in free software; and Codec2, open from the start, under M17. The line between them shapes what you can build for free and what needs hardware.
Why this page matters more than it looks. The vocoder row of the Rosetta Stone for the Digital Modes is the hinge the whole digital voice world turns on — and it is also why The Digital Cliff sounds different from one mode to the next.
A noncommercial hobby reference compiled by N6JET, gathered from public sources and shared freely for anyone interested in amateur digital voice.