Pre-processing for automatic speech recognition

Abstract

A method is provided that includes obtaining two or more microphone audio signals; analysing the two or more microphone audio signals for a defined noise type; and processing the two or more microphone audio signals based on the analysis to generate at least one audio signal suitable for automatic speech recognition. A corresponding apparatus is also provided.

1 . An apparatus comprising at least one processor and at least one memory including a computer program code, the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus at least to:
 obtain two or more microphone audio signals;   analyse the two or more microphone audio signals for a defined noise type; and   process the two or more microphone audio signals based on the analysis to generate at least one audio signal suitable for automatic speech recognition.   
     
     
         2 . The apparatus as claimed in  claim 1 , wherein the apparatus is caused to analyse the two or more microphone audio signals by being further caused to:
 determine energy estimates for the two or more microphone audio signals;   determine correlation estimates between pairs of the two or more microphone audio signals;   determine a defined noise type noise estimate based on the energy estimates for the two or more microphone audio signals and the correlation estimates between pairs of the two or more microphone audio signals; and   determine a defined noise type noise frequency threshold below which the defined noise type noise is a dominant disturbance based on the defined noise type noise estimate, the energy estimates for the two or more microphone audio signals and the correlation estimates between pairs of the two or more microphone audio signals.   
     
     
         3 . The apparatus as claimed in  claim 2 , wherein the apparatus is caused to process the two or more microphone audio signals by being further caused to:
 select, for frequency bands below the defined noise type noise frequency threshold, a lowest energy microphone audio signal of the two or more microphone audio signals; and   select, for frequency bands above the defined noise type frequency threshold, a highest energy microphone audio signal of the two or more audio signals.   
     
     
         4 . The apparatus as claimed in  claim 3 , wherein the apparatus is further caused to:
 select, for frequency bands below the defined noise type noise frequency threshold, a lowest energy microphone audio signal of the two or more microphone audio signals; and   generate, for frequency bands above the defined noise type noise frequency threshold, a filter-and-sum combination of the two or more microphone audio signals.   
     
     
         5 . The apparatus as claimed in  claim 4 , wherein the apparatus is caused to generate the filter-and-sum combination by being further caused to:
 time-align the two or more microphone audio signals; and   generate a weighted average of the time-aligned two or more microphone audio signals.   
     
     
         6 . The apparatus as claimed in  claim 4 , wherein the apparatus is caused to time-align by being further caused to:
 estimate a direction of sound arrival; and   filter the two or more microphone audio signals based on the direction of sound arrival and a microphone configuration defining the relative locations of microphones configured to capture the two or more microphone audio signals.   
     
     
         7 . The apparatus as claimed in  claim 6 , wherein the apparatus is caused to estimate a direction of sound arrival by being further caused to one of:
 estimate a direction of sound arrival based on a cross-correlation analysis of the two or more audio signals;   estimate a direction of sound arrival based on a cross-correlation analysis of the two or more audio signals when an active speech segment is detected; or   estimate a direction of sound arrival based on a cross-correlation analysis of the two or more audio signals when an active speech segment is detected and a noise level of the two or more audio signals is lower than a threshold value.   
     
     
         8 . The apparatus as claimed in  claim 1 , wherein the defined noise type comprises at least one of:
 structure borne noise;   motor noise;   actuator noise;   wind noise; or   handling noise.   
     
     
         9 . The apparatus as claimed in  claim 1 , wherein the apparatus is caused to obtain two or more microphone audio signals by being further caused to at least one of:
 receive the two or more microphone audio signals from the two or more microphones; or   retrieve the two or more microphone audio signals from memory.   
     
     
         10 . The apparatus as claimed in  claim 1 , wherein the two or more microphone audio signals are captured from at least one of:
 directional microphones;   pressure microphones; or   pressure gradient microphones.   
     
     
         11 . A method comprising:
 obtaining two or more microphone audio signals;   analysing the two or more microphone audio signals for a defined noise type; and   processing the two or more microphone audio signals based on the analysing to generate at least one audio signal suitable for automatic speech recognition.   
     
     
         12 . The method as claimed in  claim 11 , wherein analysing the two or more microphone audio signals for the defined noise type comprises:
 determining energy estimates for the two or more microphone audio signals;   determining correlation estimates between pairs of the two or more microphone audio signals;   determining a defined noise type noise estimate based on the energy estimates for the two or more microphone audio signals and the correlation estimates between pairs of the two or more microphone audio signals; and   determining a defined noise type noise frequency threshold below which the defined noise type noise is a dominant disturbance based on the defined noise type noise estimate, the energy estimates for the two or more microphone audio signals and the correlation estimates between pairs of the two or more microphone audio signals.   
     
     
         13 . The method as claimed in  claim 12 , wherein processing the two or more microphone audio signals comprises:
 selecting, for frequency bands below the defined noise type noise frequency threshold, a lowest energy microphone audio signal of the two or more microphone audio signals; and   selecting, for frequency bands above the defined noise type frequency threshold, a highest energy microphone audio signal of the two or more audio signals.   
     
     
         14 . The method as claimed in  claim 13 , wherein processing the two or more microphone audio signals comprises:
 selecting, for frequency bands below the defined noise type noise frequency threshold, a lowest energy microphone audio signal of the two or more microphone audio signals; and   generating, for frequency bands above the defined noise type noise frequency threshold, a filter-and-sum combination of the two or more microphone audio signals.   
     
     
         15 . The method as claimed in  claim 14 , wherein generating, for frequency bands above the defined noise type noise frequency threshold, a filter-and-sum combination of the two or more microphone audio signals comprises:
 time-aligning the two or more microphone audio signals; and   generating a weighted average of the time-aligned two or more microphone audio signals.   
     
     
         16 . The method as claimed in  claim 14 , wherein time-aligning the two or more microphone audio signals comprises:
 estimating a direction of sound arrival; and   filtering the two or more microphone audio signals based on the direction of sound arrival and a microphone configuration defining the relative locations of microphones configured to capture the two or more microphone audio signals.   
     
     
         17 . The method as claimed in  claim 16 , wherein estimating a direction of sound arrival comprises one of:
 estimating a direction of sound arrival based on a cross-correlation analysing of the two or more audio signals;   estimating a direction of sound arrival based on a cross-correlation analysing of the two or more audio signals when an active speech segment is detected; or   estimating a direction of sound arrival based on a cross-correlation analysing of the two or more audio signals when an active speech segment is detected and a noise level of the two or more audio signals is lower than a threshold value.   
     
     
         18 . The method as claimed in  claim 11 , wherein the defined noise type comprises at least one of:
 structure borne noise;   motor noise;   actuator noise;   wind noise; or   handling noise.   
     
     
         19 . The method as claimed in  claim 11 , wherein obtaining two or more microphone audio signals comprises, at least one of:
 receiving the two or more microphone audio signals from the two or more microphones; or   retrieving the two or more microphone audio signals from memory.   
     
     
         20 . The method as claimed in  claim 11 , wherein the two or more microphone audio signals are captured from at least one directional microphone, and processing the two or more microphone audio signals based on the analysing to generate at least one audio signal suitable for automatic speech recognition comprises: filter-and-summing the two or more microphone audio signals to generate a directional audio signal.