Robust audio identification with interference cancellation
Audio distortion compensation methods to improve accuracy and efficiency of audio content identification are described. The method is also applicable to speech recognition. Methods to detect the interference from speakers and sources, and distortion to audio from environment and devices, are discussed. Additional methods to detect distortion to the content after performing search and correlation are illustrated. The causes of actual distortion at each client are measured and registered and learnt to generate rules for determining likely distortion and interference sources. The learnt rules are applied at the client, and likely distortions that are detected are compensated or heavily distorted sections are ignored at audio level or signature and feature level based on compute resources available. Further methods to subtract the likely distortions in the query at both audio level and after processing at signature and feature level are described.
1. A method for audio identification that is robust in cancelling interfering noise from a distorting or interfering audio environment, the method comprising:
processing audio signals into filter channels to generate a two dimensional (2D) representation of an audio frequency and auto-correlation delay to extract audio features from the 2D representation and to identify a primary audio source and multiple secondary audio sources for appropriate coherent segments;
grouping filter channels into channel groups each associated with a single pitch in the appropriate coherent segments;
combining the appropriate coherent segments that are related for processing by independent hidden Markov models (HMIs) to model the primary audio source and the multiple secondary audio sources; and
detecting and canceling interfering sources and attributes to generate speech fragments for the primary audio source.
2. The method of claim 1 , further comprising:
canceling out secondary audio sources that are likely interfering sources.
3. The method of claim 2 , further comprising:
identifying interfering sources by segmenting the audio signals and utilizing rules to identify both the primary audio source and the likely interfering sources.
4. The method of claim 1 , further comprising:
measuring distortion interference on prior queries by comparing reference signatures and features to matching query signatures and features; and
learning rules to compensate the distortion interference.
5. The method of claim 1 , further comprising:
generating a reconstructed audio spectrogram, transform coefficients, and new features for an identified time when an interfering source is canceled.
6. The method of claim 1 , wherein audio filtered transform coefficients are used by each independent HMM to detect likely phonemes and further processed for grammar and vocabulary pattern matching.
7. A system comprising:
a receiver configured to receive an audio signal;
a memory that stores instructions; and
one or more processors configured by the instructions to perform operations including:
processing audio signals into filter channels to generate a two dimensional (2D) representation of an audio frequency and auto-correlation delay to extract audio features from the 2D representation and to identify a primary audio source and multiple secondary audio sources for appropriate coherent segments,
grouping filter channels into channel groups each associated with a single pitch in the appropriate coherent segments,
combining the appropriate coherent segments that are related for processing by independent hidden Markov models (HMIs) to model the primary audio source and the multiple secondary audio sources, and
detecting and canceling interfering sources and attributes to generate speech fragments for the primary audio source.
8. The system of claim 7 , wherein the operations additionally include:
canceling out secondary audio sources that are likely interfering sources.
9. The system of claim 8 , wherein the operations additionally include:
identifying interfering sources by segmenting the audio signals and utilizing rules to identify both the primary audio source and the likely interfering sources.
10. The method of claim 7 , wherein the operations additionally include:
measuring distortion interference on prior queries by comparing reference signatures and features to matching query signatures and features; and
learning rules to compensate for the distortion interference.
11. The system of claim 7 , wherein the operations additionally include:
generating a reconstructed audio spectrogram, transform coefficients, and new features for an identified time when an interfering source is canceled.
12. The system of claim 7 , wherein audio filtered transform coefficients are used by each independent HMM to detect likely phonemes and further processed for grammar and vocabulary pattern matching.
13. The system of claim 7 , wherein directional samplers and microphones provide the multiple secondary audio sources.
14. A non-transitory machine-readable medium having instructions embodied thereon, which, when executed by one or more processors of a machine, cause the machine to perform operations comprising:
processing audio signals into filter channels to generate a two dimensional (2D) representation of an audio frequency and auto-correlation delay to extract audio features from the 2D representation and to identify a primary audio source and multiple secondary audio sources for appropriate coherent segments;
grouping filter channels into channel groups each associated with a single pitch in the appropriate coherent segments;
combining the appropriate coherent segments that are related for processing by independent hidden Markov models (HMIs) to model the primary audio source and the multiple secondary audio sources; and
detecting and canceling interfering sources and attributes to generate speech fragments for the primary audio source.
15. The non-transitory machine-readable medium of claim 14 , wherein the operations additionally include:
canceling out secondary audio sources that are likely interfering sources.
16. The non-transitory machine-readable medium of claim 15 , wherein the operations additionally include:
identifying interfering sources by segmenting the audio signals and utilizing rules to identify both the primary audio source and the likely interfering sources.
17. The method of claim 14 , wherein the operations additionally include:
measuring distortion interference on prior queries by comparing reference signatures and features to matching query signatures and features; and
learning rules to compensate for the distortion interference.
18. The non-transitory machine-readable medium of claim 14 , wherein the operations additionally include:
generating a reconstructed audio spectrogram, transform coefficients, and new features for an identified time when an interfering source is canceled.
19. The non-transitory machine-readable medium of claim 14 , wherein audio filtered transform coefficients are used by each independent HMI to detect likely phonemes and further processed for grammar and vocabulary pattern matching.
20. The non-transitory machine-readable medium of claim 14 , wherein directional samplers and microphones provide the multiple secondary audio sources.