Model-driven candidate sorting based on audio cues

1. A method comprising:

identifying, by an audio cue generator executing on a processing device, utterances in audio data of a digital interview, wherein the utterances each comprise a group of one or more words spoken by a candidate in the digital interview;

generating, by the audio cue generator, audio cues of the digital interview based on the identified utterances;

applying the audio cues to a prediction model to predict an achievement index for the candidate based on the audio cues; and

displaying the candidate in a list of candidates based on the achievement index, wherein the list of candidates is sorted according to the candidates' achievement index.

2. The method of claim 1 , further comprising:

performing the identifying, generating, and applying for a plurality of additional candidates to predict additional achievement indices for the plurality of additional candidates; and

sorting the candidate and the plurality of additional candidates in the list of candidates according to their respective achievement index.

3. The method of claim 1 , wherein the identifying the utterances comprises:

computing a first moving average using a first moving window of a first number of audio samples of the audio data;

computing a second moving average using a second moving window of a second number of the audio samples with gradients to detect peak transitions of the audio samples, wherein the second number is greater than the first number;

computing a voice activity indicator of the audio samples;

segmenting the audio samples into the utterances using the first moving average, the second moving average and the voice activity indicator, wherein each of the utterances comprise an utterance length; and

assigning an utterance identifier to each of the utterances.

4. The method of claim 3 , further comprising:

identifying, by the audio cue generator, gaps between the utterances in the audio data, wherein each of the gaps comprise a gap length;

assigning a gap identifier to each of the gaps; and

computing summary statistics on the gaps.

5. The method of claim 4 , further comprising:

generating an utterance vector comprising the utterance identifiers and corresponding utterance lengths;

generating a gap vector comprising the gap identifiers and corresponding gap lengths; and

computing a ratio of utterance length to gap length using the utterance vector and the gap vector.

6. The method of claim 3 , further comprising:

performing a spectral analysis of each of the identified utterances; and

performing at least one of single-word or phrase detection or interview fingerprinting based on the spectral analysis.

7. The method of claim 3 , further comprising:

performing a spectral analysis of each of the identified utterances to generate utterance time-series data; and

computing summary statistics for each window of a spectrum, wherein each window is defined by a unit step frequency, wherein the summary statistics comprise at least one of maximum, minimum, median, skew, standard deviation, mode, slope, or kurtosis.

8. The method of claim 3 , further comprising:

performing a spectral analysis of each of the identified utterances to generate utterance time-series data;

generating a spectral signature of each of the identified utterances based on the spectral analysis; and

detecting repetition of utterances based on the spectral signatures.

9. The method of claim 8 , wherein the detecting the repetition of utterances comprises:

building an utterance matrix to represent the digital interview;

interpolating an utterance spectral response to fit within a predefined resolution of columns of the utterance matrix to allow vector combination; and

computing a Euclidean norm of the utterance matrix to estimate repetition; and

generating an interview repetition score based on the Euclidean norm.

10. The method of claim 3 , wherein the generating the audio cues further comprises:

performing an utterance Power Spectrum Density (PSD) function on the identified utterances data; and

computing an utterance repetition metric based on the PSD function.

11. The method of claim 1 , wherein the generating the audio cues further comprises:

performing Fast Fourier Transform (FFT) variants on a sound spectrum of the audio data for frequency spectral analysis; and

generating frequency statistics based on the frequency spectral analysis.

12. The method of claim 1 , wherein the generating the audio cues further comprises:

performing a spectral analysis of each of the identified utterances to obtain an utterance power spectrum density; and

computing utterance power spectrum density performing at least one of single-word or phrase detection or interview fingerprinting based on the spectral analysis.

13. The method of claim 1 , further comprising converting the audio data to text, and wherein generating the audio cues further comprises analyzing the text.

14. The method of claim 13 , wherein analyzing the text comprises at least one of:

calculating a grammar score of the text;

calculating a sentiment score of the text, wherein the sentiment score represents a mood of the text based on weights given to each word or groups of words;

calculating a count of words on a black list;

calculating summary statistics on word lengths of the text or word difficulty of the text;

calculating a frequency of filler words of the text; or

calculating position-specific word clusters for predictive words.

15. A non-transitory computer readable storage medium including instructions that, when executed by a processing device of a computing system, cause the computing system to perform operations comprising:

identifying, by the processing device, utterances in audio data of a digital interview, wherein the utterances each comprise a group of one or more words spoken by a candidate in the digital interview;

generating, by the processing device, audio cues of the digital interview based on the identified utterances;

applying the audio cues to a prediction model to predict an achievement index for the candidate based on the audio cues; and

displaying, by the processing device, the candidate in a list of candidates based on the achievement index, wherein the list of candidates is sorted according to the candidates' achievement index.

16. The non-transitory computer readable storage medium of claim 15 , wherein the operations further comprise:

performing the identifying, generating, and applying for a plurality of additional candidates to predict additional achievement indices for the plurality of additional candidates; and

sorting the candidate and the plurality of additional candidates in the list of candidates according to their respective achievement index.

17. The non-transitory computer readable storage medium of claim 15 , wherein the identifying the utterances comprises:

computing a first moving average using a first moving window of a first number of audio samples of the audio data;

computing a second moving average using a second moving window of a second number of the audio samples with gradients to detect peak transitions of the audio samples, wherein the second number is greater than the first number;

computing a voice activity indicator of the audio samples;

segmenting the audio samples into the utterances using the first moving average, the second moving average and the voice activity indicator, wherein each of the utterances comprise an utterance length; and

assigning an utterance identifier to each of the utterances.

18. A computing system comprising:

a data storage device; and

a processing device, coupled to the data storage device, to execute a digital interviewing platform comprising an audio cue generator, wherein the audio cue generator is to:

identify utterances in audio data of a digital interview, wherein the utterances each comprise a group of one or more words spoken by a candidate in the digital interview;

generate audio cues of the digital interview based on the identified utterances;

apply the audio cues to a prediction model to predict an achievement index for the candidate based on the audio cues; and

display the candidate in a list of candidates based on the achievement index, wherein the list of candidates is sorted according to the candidates' achievement index.

19. The computing system of claim 18 , wherein the audio cue generator is further to:

compute a first moving average using a first moving window of a first number of audio samples of the audio data;

compute a second moving average using a second moving window of a second number of the audio samples with gradients to detect peak transitions of the audio samples, wherein the second number is greater than the first number;

compute a voice activity indicator of the audio samples;

segment the audio samples into the utterances using the first moving average, the second moving average and the voice activity indicator, wherein each of the utterances comprise an utterance length; and

assign an utterance identifier to each of the utterances.

20. The computing system of claim 19 , wherein the audio cue generator is further to:

identify gaps between the utterances in the audio data, wherein each of the gaps comprise a gap length;

assign a gap identifier to each of the gaps; and

compute summary statistics on the gaps.