Improving speech intelligibility utilizing an articulation index
Background noise is modeled from an input signal comprising a desired signal and a plurality of undesired signals. At least one of the signals that comprise the input is processed to generate a signal-to-noise ratio. An articulation index is generated for the at least one of the signals that is processed. A spectrum of a speech segment is generated to improve intelligibility and quality of the speech segment based on the articulation index. A shaping logic may adjust the spectrum of the speech segment based on a comparison of the articulation index to a plurality of predetermined thresholds. Modeling of the background noise comprises modeling a tilt of the background noise.
1. A system that improves speech intelligibility and speech quality of a speech segment comprising:
one or more processors or circuits including:
a background noise processor programmed to detect and model a background noise from an input comprising a plurality of signals;
a signal-to-noise processor programmed to approximate a signal-to-noise ratio of at least one of the plurality of signals; and
an articulation index processor programmed to approximate an articulation index of the at least one of the plurality of signals processed by the signal-to-noise processor;
wherein a shaping processor adjusts a spectrum of the at least one of the plurality of signals, based on a comparison of the articulation index to a plurality of predetermined thresholds.
2. The system of claim 1 wherein the shaping processor is programmed to adjust the spectrum of a speech segment to improve an intelligibility and quality of the speech segment.
3. The system of claim 2 where the shaping processor, the articulation index processor, the signal-to-noise processor, and the background noise processor, comprises a unitary device.
4. The system of claim 1 where the articulation index processor, the signal-to-noise processor, and the background noise processor, comprises a unitary device.
5. The system of claim 1 where the model comprises fitting a line to the detected background noise.
6. The system of claim 5 where the model comprises approximating an inverse linear relationship.
7. The system of claim 1 where the model comprises approximating an inverse linear relationship.
8. The system of claim 1 where the model comprises approximating a non-linear relationship.
9. A non-transitory computer readable medium having software that improves a speech intelligibility and speech quality that models a speech segment based on a detected background comprising:
a modeling logic that represents the background noise detected from an input signal comprising a desired signal and a plurality of undesired signals;
a signal-to-noise logic that approximate a signal-to-noise ratio of at least one of the signals that comprise the input signal;
an articulation logic that approximates an articulation index of the at least one of the signals that is processed by the signal-to-noise processor; and
shaping logic that adjusts the spectrum of the speech segment to improve an intelligibility and quality of the speech segment, wherein the articulation index measures the intelligibility of the speech segment, and the shaping logic adjusts the spectrum of the speech segment based on a comparison of the articulation index to a plurality of predetermined thresholds.
10. The non-transitory computer readable medium of claim 9 , further comprising a memory linked to a plurality of articulation indexes.
11. The non-transitory computer readable medium of claim 10 , where at least some of the plurality of articulation indexes are customized to an interior of an enclosure.
12. The non-transitory computer readable medium of claim 9 , wherein the modeling logic includes voice activity detection that identifies whether the input signal comprises a speech signal, an unvoiced signal, or a background noise.
13. The non-transitory computer readable medium of claim 12 , wherein the desired signal comprises the speech signal and the undesired signals comprises at least one of the unvoiced signal or the background noise.
14. The non-transitory computer readable medium of claim 9 , wherein the modeling logic comprises a coherence estimation that estimates a spectral coherence between at least one of the undesired signals and the desired signal by quantifying the quality of interference between at least one of the undesired signals and the desired signal.
15. The non-transitory computer readable medium of claim 9 , wherein the articulation index measures the intelligibility of the speech segment, and the shaping logic adjusts the spectrum of the speech signal based on a comparison of the articulation index to a plurality of predetermined thresholds.
16. A method for improving speech comprising:
in one or more computing devices:
modeling background noise from an input signal comprising a desired signal and a plurality of undesired signals;
processing at least one of the signals that comprise the input to generate a signal-to-noise ratio;
generating an articulation index of the at least one of the signals that is processed; and
adjusting a spectrum of a speech segment to improve an intelligibility and quality of the speech segment based on the articulation index;
wherein a shaping logic adjusts the spectrum of the speech segment based on a comparison of the articulation index to a plurality of predetermined thresholds.
17. The method of claim 16 wherein the modeling background noise comprises modeling a tilt of the background noise.
18. The method of claim 16 wherein a signal-to-noise processor approximates the signal-to-noise ratio and an articulation index processor is programmed to approximate an articulation index.
19. The method of claim 16 wherein the articulation index measures intelligibility of the speech segment.
20. The method of claim 19 wherein the spectrum of the speech signal is adjusted based on a comparison of the articulation index to a plurality of predetermined thresholds.