System and method for voice morphing
A system and method for masking an identity of a speaker of natural language speech, such as speech clips to be labeled by humans in a system generating voice transcriptions for training an automatic speech recognition model. The natural language speech is morphed prior to being presented to the human for labeling. In one embodiment, morphing comprises pitch shifting the speech randomly either up or down, then frequency shifting the speech, then pitch shifting the speech in a direction opposite the first pitch shift.
1. A method of morphing natural language speech, comprising: receiving a natural language speech clip;
first pitch shifting the received natural language speech clip in a first direction by a first percentage;
frequency shifting the first pitch shifted speech clip by a frequency shift amount; and
second pitch shifting the frequency shifted speech clip in a second direction that is opposite to the first direction by a second percentage, to yield a morphed speech clip.
2. The method of claim 1 , where the frequency shift amount of the frequency shifting is determined randomly.
3. The method of claim 1 , wherein the frequency shifting is in an up direction.
4. The method of claim 1 , where the frequency shift amount is in a range of 100-280 Hz.
5. The method of claim 1 , where the frequency shifting comprises adding non-zero amplitude spectral lines at the low end of the frequency spectrum.
6. The method of claim 1 , where the first direction of the first pitch shifting is determined randomly.
7. The method of claim 1 , wherein the first percentage of the pitch shifting is in a range of 15 to 200% of the received speech clip.
8. The method of claim 1 , wherein the first percentage of the first pitch shift is varied randomly within a range of 15 to 200%.
9. The method of claim 1 , wherein the second percentage is varied randomly within 0 to 10% of the opposite of the first pitch shift.
10. The method of claim 1 , further comprising a permission level implemented by a data annotator tool that determines whether unmorphed data is available to a human user.
11. The method of claim 1 , wherein the first direction is determined for each natural language clip received.
12. The method of claim 1 , wherein the first percentage is determined for each natural language clip received.
13. The method of claim 1 , wherein the frequency shift amount is determined for each natural language clip received.
14. The method of claim 1 , wherein the second percentage is varied randomly within 0 to 10% of the opposite of the first pitch shift for each natural language clip received.