Dynamic acoustic model for vehicle
A vehicle voice processor includes a processing device and a data storage medium. The processing device is programmed to receive identification information from a wearable device, identify a speaker from the identification information, identify a dialect associated with the speaker from the identification information, select a predetermined acoustic model, and adjust the predetermined acoustic model based at least in part on the dialect identified.
1. A vehicle voice processor comprising a processing device and a data storage medium, wherein the processing device is programmed to:
receive identification information from a wearable device;
identify an individual speaker from the identification information;
identify a dialect associated with the individual speaker;
select a predetermined acoustic model based on the identified individual speaker; and
then dynamically and in real-time adjust the predetermined acoustic model thereby generating a calibrated acoustic model, based at least in part on the dialect associated with the identified individual speaker, by reweighting feature vectors in the predetermined acoustic model in accordance with the dialect associated with the individual speaker.
2. The vehicle voice processor of claim 1 , wherein the processing device is programmed to apply the calibrated acoustic model to an acoustic signal upon dynamically determining the calibrated acoustic model.
3. The vehicle voice processor of claim 1 , wherein the processing device is programmed to output a vehicle command based at least in part on the acoustic signal and the calibrated acoustic model.
4. The vehicle voice processor of claim 1 , wherein adjusting the predetermined acoustic model includes selecting a voice feature from a plurality of voice features.
5. The vehicle voice processor of claim 4 , wherein adjusting the predetermined acoustic model includes adjusting a weight associated with the selected voice feature.
6. The vehicle voice processor of claim 1 , wherein adjusting the predetermined acoustic model includes adjusting a weight associated with at least one of a plurality of voice features.
7. The vehicle voice processor of claim 6 , wherein each of the plurality of voice features is associated with a phoneme.
8. The vehicle voice processor of claim 1 , further comprising a communication device programmed to pair with the wearable device.
9. The vehicle voice processor of claim 1 , further comprising a microphone configured to receive an analog acoustic signal.
10. A method comprising:
receiving identification information from a wearable device;
identifying an individual speaker from the identification information;
identifying a dialect associated with the individual speaker;
selecting a predetermined acoustic model; and
then dynamically and in real-time adjusting the predetermined acoustic model thereby generating a calibrated acoustic model, based at least in part on the dialect associated with the identified individual speaker, by reweighting feature vectors in the predetermined acoustic model in accordance with the dialect associated with the individual speaker.
11. The method of claim 10 , further comprising applying the calibrated acoustic model to an acoustic signal upon dynamically determining the calibrated acoustic model.
12. The method of claim 10 , outputting a vehicle command based at least in part on the acoustic signal and the calibrated acoustic model.
13. The method of claim 10 , wherein adjusting the predetermined acoustic model includes selecting a voice feature from a plurality of voice features.
14. The method of claim 10 , wherein adjusting the predetermined acoustic model includes adjusting a weight associated with at least one of a plurality of voice features.
15. The method of claim 14 , wherein each of the plurality of voice features is associated with a phoneme.
16. A vehicle system comprising:
a communication device programmed to pair with the wearable device;
a microphone configured to receive an acoustic signal; and
a voice processor programmed to:
receive identification information from the wearable device,
identify an individual speaker from the identification information,
identify a dialect associated with the individual speaker,
select a predetermined acoustic model based on the identified individual speaker, and
then dynamically and in real-time adjust the predetermined acoustic model thereby generating a calibrated acoustic model, based at least in part on the dialect associated with the identified individual speaker, by reweighting feature vectors in the predetermined acoustic model in accordance with the dialect associated with the individual speaker.