Godot GDextension allowing to have speech to text in your game using vosk
@onready var vosk := $Vosk
...
var effect : AudioEffectCapture # See AudioEffect in docs
func _ready() -> void
# vosk-model-small-fr-0.22 is an audio model you can download
# on vosk's website, you can replace it by the langage you want
#
# the 2nd parameters is a JSON string that allows you to limit the
# vocabulary that model will recognize, it decrease the chance of ambiguity
# and increase the accuracy
vosk.init("bin/vosk-model-small-fr-0.22", '["foo", "bar"]')
func _process(delta: float) -> void:
if !effect.can_get_buffer(buffer_length_frames / 2):
return
var data = effect.get_buffer(effect.get_frames_available())
#
# basically accept_wave_form is the method to feed it with new audio data
# (it works with stream and is able to detect change from silence to 'speaking'
# and will continue buffering as long it's not again silence"
# so you're supposed to only give it the new data since last time you called it
# not feed it everytime the whole data.
#
# it returns true when it's again silence and it has a final result
if accept_wave_form_stereo_float(data):
print(vosk.result())
else:
# if it's not yet finalized, you still can get the partial result
# it can be useful if you want to have result ASAP and you can accept a lower accuracy
var partial_result = JSON.parse_string(vosk.partial_result())["partial"]
This extension use a third party library (libvosk.so)
to do so we had to add in SConstruct the two lines
env.Append(LIBPATH=["bin/"])
env.Append(LIBS=["libvosk"])
for exporting things correctly you also need the [dependencies] section in your .gdextension file
[dependencies]
linux.debug.x86_64 = {
"res://bin/libvosk.so" : ""
}
linux.release.x86_64 = {
"res://bin/libvosk.so" : ""
}
This project is a simpler wrapper around libvosk
you can check this awesome project there