mat

M17 Audio Tools

Introduction

This is the machinery used to make the 3200 bps Codec2 *.dat files for mspot, an M17-only hot-spot/repeater. You can use this to create your own mspot audio files, or you can use it to create Codec2 dat file from any wav file.

The *dat files produced are binary files containing just the C2_3200 encoded data. That means that there are 16 byes of data for every M17 voice-only 40 millisecond frame. The size of the *.dat files produced will always be a multiple of 16. but they can be an odd number because in fact C2_3200 encodes 20 milliseconds of audio into 8 bytes. The M17 voice-only protocol uses two of these for each 40 millisecond frame.

Building the tools

1. Make sure you have the needed packages: sudo apt install git build-essential
2. Download this repo: git clone https://github.com/n7tae/mat
3. Move to the tools folder: cd mat/tools
4. Build the tools: make

This will build three programs:

• c2enc is used to create a Codec2 dat file from a monaural, 16-bit, little-endian, 8000 Hz, raw audio file.
• c2dec is used to create a raw audio file from a Codec2 dat file.
• index is used to produce a list and index file from the raw audio file.

All three programs are used via the command line and require parameters to do their job. If you don't supply any parameters they will print a usage message, but for everything described here you won't be using any of these tools directly.

Finally, move back to the repo's base folder to get ready for the next step: cd ..

Make the mspot dat files

You'll need to do this on a debian-based computer that can play audio. It's a good idea to go through this even if you are going to provide you own wav files, just to see how everything works. But first, we need some more packages:

sudo apt install espeak alsa-utils sox

If you want to use Piper to generate more realistic-sounding voices:

• Install Python: ``sudo apt install python3`
• Create a Python virtual environment: python3 -m venv .venv/
• Activate it: source .venv/bin/activate

From the repo's base directory:

1. Create a directory for all your files and go there: mkdir mspot && cd mspot
2. Make the wav files: ../speakwav (for espeak) or ../speakwav-piper (for Piper)
3. Then do the rest of the processing: ../makedat (for espeak) or ../makedat 150 (for most Piper voices)

Step 2 will create the wav files using espeak. You can hear what each wav file sounds like, for example: aplay speak.wav

Step 3, the makedat script, does all the heavy lifting:

1. The wav files are converted to raw files.
2. The raw files are analyzed. A preliminary index file is generated showing where quiet gaps are and this is used to trim the leading and trailing quiet regions of the raw file.
3. Then, in blocks of 160 16-bit values, the maximum absolute value is found and listed by block index in the list files. These list files are then used to generate the final index files.
4. The raw wave files are encoded into dat files by c2enc.

Please note that there are no usage messages for either the speakwav or the makedat script, but makedat will take exactly one optional parameter that will be covered in the next section.

In the end, only the dat files and speak.index is needed by mspot. speak.dat is the only file that mspot slices up to produce either your repeater callsign or a destination callsign dat.

Make your own voice files

You need wav files, or other files that sox can translate to raw files, or, you can just supply monaural, 16-bit, little-endian, 8000 Hz raw files. The voice contents of these files are listed in the speakwav script in the quoted string used to create each file. This is your chance to record voice files in a different language! Filenames are chosen to be descriptive and you must use those exact names for your files. The best way to make your voice files is with an audio tool like audacity or garageband using a decent microphone. Using these tools you can easily edit and enhance the recording for the best possible playback on mspot.

Once you have recorded all the needed files, you can move them to a new folder and process them with the makedat script just like in the previous section. The speak.raw file is the most critical to get right. The index program must be able to find exactly 67 words in this file. They will be indexed with 0 through 66. If index is not finding 67 words, first, listen to it to make sure you have recorded all 67 words: space, A, B, C, D, E, F, G, H, I, J, K, L, M, N, O, P, Q, R, S, T, U, V, W, X, Y, zee, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, dash, slash, dot, alpha, bravo, Charlie, delta, echo, foxtrot, golf, hotel, India, juliette, kilo, Lima, Mike, november, Oscar, papa, Quebec, romeo, Sierra, tango, uniform, Victor, whiskey, X-ray, yankee, zulu, M17.

You should be listening to the dat file. If you're in your new folder: ../tools/c2dec 3200 speak.dat - | aplay -f s16_le

If the recording volume was too low or too high, you'll have to re-record. The speak.list file should show values no greater than about 15k-25k and gaps between words should ideally have values below 10. However if the noise floor between words is larger, you can try reprocessing the raw files: ../makedat <threshold> where <threshold> is a value slightly larger than the values seen between words in your speak.list file. The object here: You want index to find exactly 67 words. Using too-small of a threshold will find too few words while using a too-large threshold might find too many words. In general, you want the smallest threshold that finds exactly 67 words.

Once you are happy with the outcome, move all your dat files and speak.index to your mspot system. Then, make sure mspot.ini points to that location. Don't put them the the mspot/audio folder because you won't be able to update your mspot repo.

Keep up!

The mspot program is new. It will probably add new messages with time and you will have to record and encode some new files when that happens. It is even possible, perhaps even likely, that the speak file will add new words.

Copyright

This use the GNU General Public License Version #2 as described in the LICENSE file.

Copyright (c) 2025,2026 by Thomas A. Early, N7TAE