US 12,136,428 B1
Audio watermarking
Ameya Agaskar, Bedford, MA (US); and Sumit Garg, Acton, MA (US)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Sep. 30, 2021, as Appl. No. 17/490,271.
Claims priority of provisional application 63/241,270, filed on Sep. 7, 2021.
Int. Cl. G10L 19/018 (2013.01); G10L 15/22 (2006.01)
CPC G10L 19/018 (2013.01) [G10L 15/22 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method, the method comprising:
receiving first audio data;
determining that a wakeword is represented in a first portion of the first audio data;
determining to insert an audio watermark in the first portion of the first audio data;
determining first feature vector data corresponding to the first audio data, the first feature vector data including:
a first power value associated with the first audio data and a first frequency range, and
a second power value associated with the first audio data and a second frequency range;
determining a first coefficient value associated with the first frequency range, wherein the first coefficient value is determined based on the second frequency range;
determining, using the first feature vector data and the first coefficient value, a first mask value representing a highest energy associated with the first frequency range;
determining, using the first feature vector data, a second mask value representing a highest energy associated with the second frequency range;
determining that the first mask value is less than the second mask value;
determining, using the first mask value, a first value corresponding to a first portion of the audio watermark; and
generating, using the first value, a second portion of second audio data by embedding the first portion of the audio watermark in the first portion of the first audio data.