Skip to content

ASL Text-to-Speech

ASL provides the asl-tts command. This is a text to speech speaker for ASL3. It uses the piper open source TTS system that is not depending on cloud providers and has opensource voice models for dozens of languages.

Basic Operation

asl-tts uses the piper-tts text to speech engine to generate ulaw files from text so that Asterisk/apt_rpt can speak any set of arbitrary text without needing to have a sound file installed for the word necessary. By default, asl-tts will cause app_rpt to immediately speak whatever text was specified.

asl-tts has a time delay

Running the piper-tts system takes a few seconds to compile the voice into the sound file. Larger text blocks may take tens of seconds to compile. asl-tts is not appropriate for time-sensitive communications such as speaking the current time.

The command format is as follows: 

asl-tts -n NODE -t 'TEXT' [ -v VOICE ] [ -f FILE ]

The general use case is to speak to a node immediately. This this case speaking the text "Good morning" to node 63001 would look like this:

sudo asl-tts -n 63001 -t 'Good morning'

The TTS engine will start, create the audio, and execute the playback on the node. The playback follows normal app_rpt rules and will be played back as soon as its turn arrives. The command will not preempt or "talk over" other traffic on the node.

Keep in mind the differences between quoting in a shell command. In general, the " and ' work the same by the double-quote will allow variable substitution while the single quote will not. This can be useful if you want to create text based on a system condition. For example:

TODAY=$(date +"%A, %B %d, %Y")
sudo asl-tts -n 63001 -t "Today is ${TODAY}"

Obviously the above example is a bit contrived as the date command doesn't generate nice words such as "first", "seventeenth", etc. However it demonstrates the flexibility of the tool. A script could be used to populate $TODAY with natural words. For a hypothetical example:

TODAY=$(/path/to/today-script)
sudo asl-tts -n 63001 -t "Today is ${TODAY}"

More complex message can be created with additional shell coding. As an example here's how the current public IP address of the system could be obtained and spoken:

IPADDR=$(wget -q -O - checkip.dyndns.com | grep -Po "[\d\.]+" | sed 's/\./ dot /g')
MSG="The node public IP is ${IPADDR}"
sudo asl-tts -n 63001 -t ${MSG}

A bash shell variable can generally hold up to a 2 Mb-sized string so any reasonably-formatted text can be stored in a variable and then played with the -t command. Keep in mind that the text or the variable contents must not contain the same quote type used in the command without escaping it with a backslash. For example "Nested \" mark".

Temporary File Cleanups

Note that temporary files are written to directories named /tmp/asl-tts and are not automatically cleaned up because it's impossible to know when Asterisk will actually need/speak the file contents. If this becomes a space problem, put in a systemd timer unit or a cron job to delete old files in /tmp/asl-tts. The command can simply be rm -rf /tmp/asl-tts*.

Creating a Sound File for Later

To create the text for other usage, such as cron jobs, use the "-f FILE" option to specify where the file should be created. Do not use an extension; the file will be automatically appended with the .ul postfix.

For example: 

AUD_TEXT="The club net will begin in five minutes!"
asl-tts -n 460181 -t "${AUD_TEXT}" -f /tmp/netstartsin5

This will create a sound file /tmp/netstartsin5.ul that can be used later with an asterisk playback such as: 

asterisk -rx 'rpt playback 63001 /tmp/netstartsin5

Note that -n is required but irrelevant to this use case.

Voices

By default, the package provides on voice "Amy" from https://github.com/rhasspy/piper/blob/master/VOICES.md. This voice is the closest voice to the default Asterisk and app_rpt sounds. It is possible to download other voices, for any language, and store the .onnx and .onxx.json files in /var/lib/piper-tts and then use them with the -v option. For example, to add the English en_GB voice alan from the voices repository:

  1. Download the .onnx and .onnx.json file to /var/lib/piper-tts

  2. Specify -v en_GB-alan-low.onnx on the command line to asl-tts. As an example:

asl-tts -n 63001 -v en_GB-alan-low.onnx -t "Good morning"

Note, since all files are squashed down to 8K uLaw format, there is no value in the "medium" or "high" quality models. Always use the low quality model.

Reporting Bugs

Report bugs to https://github.com/AllStarLink/asl3-tts/issues