Improving recognition accuracy never hurts. Many people even have to tweak something to get a good experience. Here we show you what you can do.
A quiet room will always be better than a noisy one. A good microphone can mitigate this, but sometimes a $0 job will do. Move fans and air filters, rotate your sitting position such that air conditioning is not pointing toward the mic, close a door (and subsitute by taking more breaks for health and ventilation). Move your PC from above desk to below desk, move your PC speakers behind your mic, etc.
Make sure that you have a good microphone and sound card, as speech recognition software requires better hardware than is commonly built in to computers. Have a look at the hardware page for recommendations and correct setup.
Ensure that your voice is not tired, which will lead to poorer pronunciation.
Controlling your computer by voice can be frustrating if your accent isn’t recognized well; non-native speakers are especially prone to issues. While Talon’s engine is improving with each version, in the meantime we can adapt our pronunciation.
https://rachelsenglish.com/ is an excellent resource. It features detailed videos and discussion of the mechanics behind the various sounds. At the risk of saying the obvious: you shouldn’t try to change your accent if you don’t want to.
If only a few specific commands give you trouble, change those command words! Commands are almost exclusively mapped in
.talon files: search for all occurrences of those words, and replace them with a word of your choosing: select one that is easy to pronounce, is not too similar to anything else, and is recognized well for you.
Some important commands are specified in
.py files, like the alphabet. Changing it works the same way, but you need to be more careful about what you modify. If you are not a programmer, ask in the Talon Slack
#help channel if you are unsure!
Example: If the
tab close does not work for you (but
tab open etc. do), try out
tab wipe or
tab quit instead.
zip (letter Z) is always understood as
sit (letter I), change
ivy, or alternatively change
Talon lets you configure how long it will wait after you stop speaking before trying to interpret your command. As of March 2021, the default is 0.3 seconds; earlier versions had a default of 0.15 seconds. If you find that Talon is frequently cutting you off mid-command, you may want to try one or more of the following:
Prepare your command in full before voicing it. This will help you speak fluently.
Speak faster, especially after consonants like “p” that involve stopping airflow. In utterances like “stop it”, there is a natural pause between “stop” and “it”. Some people have just slightly longer stops than others, but can avoid it when preparing to talk fluently.
Increase Talon’s wait time. This has the disadvantage of making all commands react slower, but it can eliminate the cutoffs. Do this by adding a
settings.talon file to your user directory with the following content:
# minimum silence time (in seconds) before speech is cut off, default 0.3
speech.timeout = 0.4
Start with the value of 0.4, just slightly higher than the default, and increase only if necessary.
This will not help you today, but instead help to improve the talon engine for all of us in the future. You can do this by providing speech (and noise) samples:
Speakers of non-rhotic dialectics of English (i.e. your ‘r’s don’t sound like an American’s) may find it helpful to change the alphabet. The following is a list of alternatives users mentioned on Slack.