0.2.0

Tacspeak - speech recognition for gaming

v0.2.0

This release DOES include a pre-trained model download, but it's EXPERIMENTAL!
If you're not keen on using an EXPERIMENTAL model, please download / use the pre-trained model from the 0.1 release.
download model from 0.1 release

Highlights and useful info:

YouTube video - v0.2 key changes, model & testing info

Model results should go into issue #23, link here

1. EXPERIMENTAL new kaldi model, finetuned from the base model, with ~23hrs of Ready or Not commands
   • You can use the base (0.1) model or the new EXPERIMENTAL model with Tacspeak v0.2.
   • You should only use the new EXPERIMENTAL model if you're willing to test both the new model and base model.
2. New --test_model command line argument (& more) used to test models against retained audio + metadata
3. Changes to user_settings.py
   • NoiseSink rule (enable / disable in user_settings.py). Particularly useful for the new model as it's more sensitive to noise.
   • listen_key_padding_end_ms_max and related in user_settings.py to fix audio being cut-short
   • retain_dir and related in user_settings.py to retain audio + metadata of recognitions
4. Minor improvements to Ready or Not grammar module, should recognise "gold ... command" more accurately

Full Changelog: 0.1.5...0.2.0

0.1.5

Tacspeak - speech recognition for gaming

v0.1.5 - minor update

This release doesn't include a pre-trained model download, but a pre-trained model is required to run Tacspeak!
Please download / use the pre-trained model from the 0.1 release.
download model from 0.1 release

Highlights and useful info:

• Fix deploy flashbang on ground
• Remove "belay" (cancel held order)
• Added "em" as alternative for "them", e.g. "restrain em"
• updated readme - YT video link how to use & change settings
• update to dragonfly 1.0.0-rc2-dev105
  • fixed audio during "cold mic" being prepended to retained audio (if enabled) when using global toggle mode (I don't think this was affecting speech recognition, but it's fixed now regardless)
• added powershell scripts to aid with cleaning retained audio (if enabled)
• .gitignore "retain/" folder, which is used to store audio + metadata if setting is enabled

Full Changelog: 0.1.4...0.1.5

0.1.4

Tacspeak - speech recognition for gaming

v0.1.4 - minor update

This release doesn't include a pre-trained model download, but a pre-trained model is required to run Tacspeak!
Please download / use the pre-trained model from the 0.1 release.
download model from 0.1 release

Highlights and useful info:

• update of back-end audio library
• update of build process

Full Changelog: 0.1.3...0.1.4

0.1.3

Tacspeak - speech recognition for gaming

v0.1.3 - minor update

This release doesn't include a pre-trained model download, but a pre-trained model is required to run Tacspeak!
Please download / use the pre-trained model from the 0.1 release.
download model from 0.1 release

Highlights and useful info:

• added "remove the wedge"!
• fixed close door
• team move & cover - removed "my front" and "forward"
• team halt - added "stop", "stop position", "stop movement"
• npc (civilian / suspect) move - removed "to"
• added optional "formation" suffix to diamond and wedge formations
• changes to individual team member options, but commands remain disabled until keybinds are available in-game
• added wand as equivalent for mirror
• added block as equivalent for wedge
• added "use the (wand | mirror)"
• changed leader breach + nade to include "wait for (my | me to)"
• updated readme

What's Changed

• ron - changed leader breach + nade to include "wait for (my | me to)" by @jwebmeister in #6
• RoN - Remove the Wedge! and other grammar module changes by @jwebmeister in #7

Full Changelog: 0.1.2...0.1.3

0.1.2

Tacspeak - speech recognition for gaming

v0.1.2 - minor update

This release doesn't include a pre-trained model download, but a pre-trained model is required to run Tacspeak!
Please download / use the pre-trained model from the 0.1 release.
download model from 0.1 release

Highlights and useful info:

Trapped doors

Example spoken commands (not exhaustive):

• "blue team disarm that door"
• "red team wedge that trapped door" (for a trapped door)
• "gold cover that trapped door" (for a trapped door)
• "gold cover the door" (for a un-trapped door)

The command keys will shift appropriately for the affected commands (wedge, cover, open / close door) if the user says it's a "trapped" door, or it will default to the command keys for an non-trapped door if the user does not say "trapped".

Leader breach and leader grenade

Example spoken commands (not exhaustive):

• "blue team lead will breach clear it"
• "lead will open door use flashbangs and clear it"
• "wait for my breach gas and clear"
• "red team c two wait for my flash clear it"
• "gold team kick the door lead will fourty mil clear it"

Removed other options that were cause issues with speech recognition accuracy.

On my command = hold for my command, without the hold

The word "hold" was causing issues with speech recognition accuracy. Options for holding a commands are now "on my (mark | order | command)". You can of course still modify this to whatever you want.

DEBUG_HEAVY_DUMP_GRAMMAR in user_settings

This can be used to dump out all the possible options in the ron grammar module into a txt file. I suggest leaving it as False even when troubleshooting / debugging because it can be expensive - in a previous iteration (shouldn't be the case in release version) the triggered functions chewed up 32GB of RAM and locked up the application.

Commits:

• ron - modify the door options cmd keys pressed if the user says it's "trapped" door or doesn't specify (or says "null" door)
• ron - added "disarm" (the door)
• ron - removed "hold" from "hold" command (ironic) - to improve recognition accuracy - can still use "on my command"
• Added DEBUG_HEAVY_DUMP_GRAMMAR to user_settings; updated ron to use it
• ron changed lead breach and lead grenade
• improved debug_grammar dump of possible commands
• ron - added auto stack (but there are issues in-game), removed "zip tie" "cuff" "zap" "shock" and a bunch of melee target specific commands (to improve recognition accuracy)
• ron - separated try DEBUG_MODE and try DEBUG_HEAVY_DUMP_GRAMMAR
• fixed exception handling
• show version on load. updated version to 0.1.2

0.1.1

Tacspeak - speech recognition for gaming

v0.1.1 - minor update

This release doesn't include a pre-trained model download, but a pre-trained model is required to run Tacspeak!
Please download / use the pre-trained model from the 0.1 release.
download model from 0.1 release

Changes:

• added optional arg "--print_mic_list" that will help with setting "input_device_index" (if required)
• Ready or Not grammar module - added alt word "secure" for search the room. disabled individual team member commands (waiting on keybinds)

---

Additional info:

If you're having issues with audio devices, and setting the default in Windows doesn't help, you can now run ./tacspeak.exe --print_mic_list in Powershell or command prompt to help with setting input_device_index.

• This will list all of the audio devices found on your system, and can be useful for figuring out the correct index number for the input_device_index setting in ./tacspeak/user_settings.py.
• A far easier option to try first is to set the correct default recording device in Windows Sound Settings.

Also related info - the underlying model that Tacspeak currently uses is based on "16-bit Signed Integer PCM 1-channel 16kHz" audio. Tacspeak tries to convert the incoming audio from your device to this format, but if it's too much for a single CPU core to convert in real-time it may fall over.

• I've had no issues using Tacspeak with a 48kHz, 16-bit, 2-channel microphone array and also using a Rode AI-1 and Podmic at 48kHz, 24-bit, 1-channel.
• If, for example, your device is recording at 144kHz, or something a single core on your CPU can't handle, it will likely display errors in the console.

0.1

Tacspeak - speech recognition for gaming

v0.1 - initial release

Changes:

• added Ready or Not (for game version 1.0) grammar module, see ./tacspeak/grammar/_readyornot.py
• added user settings, see ./tacspeak/user_settings.py
• release includes model, Kaldi Active Grammar, originally from daanzu/kaldi-active-grammar/releases/tag/v3.1.0

---

Additional info:

Useful user settings

• User settings, see ./tacspeak/user_settings.py
• listen_key and listen_key_toggle
  • listen_key=0x05
    • 0x05 = mouse thumb button 1.
    • 0x10 = Shift key.
    • See here for more info.
  • listen_key_toggle=-1
    • Recommended is 0 or -1.
    • 0 for toggle mode off, listen only while key is pressed; must release key for the command to be recognised.
    • 1 for toggle mode on, key press toggles listening on/off; must toggle off for the command to be recognised.
    • 2 for global toggle mode on, key press toggles listening on/off, but it uses Voice Activity Detector (VAD) to detect end of speech and recognise commands so you don't have to toggle off to recognise commands.
    • -1 for toggle mode off + priority, listen only while key is pressed, except always listen for priority grammar ("freeze!") even when key is not pressed.
    • None always listening; similar to global toggle on, uses Voice Activity Detector (VAD) to detect end of speech and recognise commands.

Useful grammar module settings and notes

• Ready or Not, see ./tacspeak/grammar/_readyornot.py
  • ingame_key_bindings should be changed if your keybindings are different to the games default
  • see ./tacspeak/grammar/_readyornot.py to understand what commands are available. It should be mostly intuitive, and/or matches the in-game command menu, but your mileage may vary.