System and method for voice morphing in a data annotator tool
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. Labeling the morphed speech comprises at least one or more of transcribing the morphed speech, identifying a gender of the speaker, identifying an accent of the speaker, and identifying a noise type of the morphed speech.
1. A system for transcribing natural language speech, the system comprising: a computer implementing a data annotator tool that performs:
receiving an audio clip comprising the natural language speech from a server;
morphing the audio clip to a morphed audio clip where the audio clip is pitch shifted, frequency shifted, and pitch shifted a second time;
playing the morphed audio clip for a human being;
receiving a transcription input from the human being for the morphed audio clip; and
providing the transcription input to a memory, wherein the data annotator tool further comprises:
a first UI area that allows the human being to play the morphed audio clip;
a second UI area that allows the human being to enter the transcription input of the morphed audio clip;
a third UI area that allows the human being to enter a gender input for the morphed audio clip;
a fourth UI area that allows the human being to enter an accent input for the morphed audio clip; and
a fifth UI area that allows the human being to enter a noise input of the morphed audio clip.
2. The system of claim 1 , wherein the morphing comprises:
first pitch shifting the received audio clip;
frequency shifting the pitch shifted speech clip; and
pitch shifting the frequency shifted speech clip in a direction opposite to the first pitch shift.
3. The system of claim 1 , wherein the data annotator tool further performs:
receiving a gender input from the human being representing the human being's identification of a gender of a speaker in the morphed audio clip; and
sending the gender input for the morphed audio clip to the server.
4. The system of claim 1 , wherein the data annotator tool further performs:
receiving an accent input from the human being, representing the human being's identification of an accent present in the morphed audio clip; and
sending the accent input for the morphed audio clip to the server.
5. The system of claim 1 , wherein the data annotator tool further performs:
receiving a noise input from the human being representing the human being's identification of noise effects present in the morphed audio clip; and
sending the noise input for the morphed audio clip to the server.
6. The system of claim 5 , wherein the noise input received from the human being for the morphed audio clip comprises one or more of: wind, voice/babble, and reverb.
7. The system of claim 1 , wherein the data annotator tool further performs
displaying parameters of morphing rules specific to the morphed audio clip.
8. The system of claim 1 , further comprising:
receiving a second audio clip from the server; and
morphing the second audio clip to a second morphed audio clip where the second audio clip is pitch shifted, frequency shifted, and pitch shifted,
wherein the morphed audio clip and the second morphed audio clip are morphed using different morphing techniques from each other.
9. The system of claim 1 , wherein the data annotator tool receives an audio clip and a second audio clip, the audio clip and the second audio clip being from the same speaker, and wherein a first morphing technique is used to create the morphed audio clip from the audio clip and a second morphing technique is used to create a second input morphed audio clip from the second audio clip.
10. A system of transcribing natural language speech, the system comprising:
a computer that performs:
receiving an audio clip comprising the natural language speech, and
morphing the audio clip to a morphed audio clip, where the audio clip is pitch shifted, frequency shifted, and pitch shifted; and
a data annotator tool that performs:
receiving the morphed audio clip,
playing the morphed audio clip for a human being,
receiving a transcription input from the human being for the morphed audio clip, and
providing the transcription input to an ASR training module for use in training a model, wherein the data annotator tool further comprises:
a first UI area that allows the human being to play the morphed audio clip;
a second UI area that allows the human being to enter the transcription input of the morphed audio clip;
a third UI area that allows the human being to enter a gender input for the morphed audio clip;
a fourth UI area that allows the human being to enter an accent input for the morphed audio clip; and
a fifth UI area that allows the human being to enter a noise input of the morphed audio clip.
11. The system of claim 10 , wherein the morphing comprises:
pitch shifting the received audio clip in a first direction;
frequency shifting the pitch shifted speech clip; and
pitch shifting the frequency shifted speech clip in a second direction opposite to the first direction.
12. The system of claim 10 , wherein the data annotator tool further performs:
Receiving the gender input from the human being representing the human being's identification of a gender of a speaker in the morphed audio clip; and
providing the gender input to the ASR training module for use in training the model.
13. The system of claim 10 , wherein the data annotator tool further performs:
receiving the accent input from the human being, representing the human being's identification of an accent present in the morphed audio clip; and
providing the accent input to the ASR training module for use in training the model.
14. The system of claim 10 , wherein the data annotator tool further performs:
receiving the noise input from the human being representing the human being's identification of noise effects present in the morphed audio clip; and
providing the noise input to the ASR training module for use in training the model.
15. The system of claim 14 , wherein the noise input received from the human being for the morphed audio clip comprises one or more of: wind, voice/babble, and reverb.
16. The system of claim 10 , wherein the data annotator tool further performs:
displaying parameters of morphing rules specific to the morphed audio clip.
17. A method for transcribing natural language speech, the method comprising, performed by a data annotator tool:
receiving an audio clip comprising the natural language speech from a server;
morphing the audio clip to a morphed audio clip where the audio clip is pitch shifted, frequency shifted, and pitch shifted,
playing the morphed audio clip for a human being;
receiving a transcription input from the human being for the morphed audio clip; and
providing the transcription input to a memory, wherein the data annotator tool further comprises:
a first UI area that allows the human being to play the morphed audio clip;
a second UI area that allows the human being to enter the transcription input of the morphed audio clip;
a third UI area that allows the human being to enter a gender input for the morphed audio clip;
a fourth UI area that allows the human being to enter an accent input for the morphed audio clip; and
a fifth UI area that allows the human being to enter a noise input of the morphed audio clip.