Skip to content

Latest commit

 

History

History
108 lines (73 loc) · 7.45 KB

readme.md

File metadata and controls

108 lines (73 loc) · 7.45 KB
Raw

WAWEnets MATLAB® Code

Implements Wideband Audio Waveform Evaluation networks or WAWEnets.

This WAWEnets implementation produces one or more speech quality or intelligibility values for each input speech signal without using reference speech signals. WAWEnets are convolutional neural networks and they have been trained using full-reference objective speech quality and speech intelligibility values.

Details can be found in the WAWEnets paper.1

If you need to cite our work, please use the following:

@INPROCEEDINGS{
   9054204,
   author={A. A. {Catellier} and S. D. {Voran}},
   booktitle={ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
   title={Wawenets: A No-Reference Convolutional Waveform-Based Approach to Estimating Narrowband and Wideband Speech Quality},
   year={2020},
   volume={},
   number={},
   pages={331-335},
}

How to run

  • Clone to local repository
  • Open using MATLAB or Octave
  • Run WAWEnet.m.

Many example calls and results are provided in demo.m.

Input can be a vector of speech samples, the name of a .wav file, or the name of a text file containing a list of .wav files.

Results are shown in the MATLAB command window, are returned in a structure, and can also be logged into an optional output text file.

Further details below (or type help WAWEnet in the MATLAB command window).

Usage

[allFileInfo, ctlInfo] = WAWEnet(inSpeech, ctlInfo)

Inputs

The only required input is inSpeech. This may be a row or column vector of speech samples or a string of the form *.wav or *.txt.

If inSpeech is a vector of speech samples the sample rate must be 16,000 smp/sec, it must contain at least 48,000 samples (3 sec.) and the nominal range for the sample values is +/- 1.0.

To best match the designed scope of WAWEnets, it should have a speech activity factor of roughly 0.5 or greater and an active speech level near 26 dB below the clipping points of +/- 1.0. (See level normalization feature below.)

If inSpeech is a string that names a .wav file, that .wav file must:

  • be uncompressed
  • have sample rate 8, 16, 24, 32, or 48k smp/sec.
  • contain at least 3 seconds of speech

To best match the designed scope of WAWEnets, the .wav file should have a speech activity factor of roughly 0.5 or greater and an active speech level near 26 dB below the clipping points of +/- 1.0. (See level normalization feature below.) The native sample rate for WAWEnets is 16 k smp/sec so files with rates 8, 24, 32, or 48k rate are converted internally before processing.

If inSpeech is a string that names a .txt file, each line should specify a .wav file that meets the .wav file requirements listed above. Each file will be processed in sequence. This will be slightly more efficient (1 or 2% reduction in run-time) than calling WAWEnet.m repeatedly.

ctlInfo is an optional input. It is a structure that provides control information to WAWEnet.m. Any or all of the five different fields may be used in any combination as follows:

ctlInfo.WAWEnetMode = M invokes the WAWEnet trained using a specific full-reference target.

  • M = 1: WAWEnet trained using WB-PESQ2 target values (Default)
  • M = 2: WAWEnet trained using POLQA3 target values
  • M = 3: WAWEnet trained using PEMO4 target values
  • M = 4: WAWEnet trained using STOI5 target values

ctlInfo.levelNormalization = L specifies internal level normalization of the speech input (vector or .wav file) to 26 dB below clipping.

  • L = 0: normalization off
  • L = 1: normalization on (Default)

ctlInfo.segmentStep = S specifies the segment step (stride) and is an integer with value 1 or greater. Default is S = 48,000. WAWEnet requires a full 3 seconds of signal to generate a result. If the input speech (vector or .wav file) is longer than 3 seconds multiple results may be produced. S specifies the number of samples to move ahead in the vector or file when extracting the next segment. The default value of 48,000 gives zero overlap between segments. Using this default any input less than 6 sec. will produce one result, based on just the first 3 sec. A 6 sec. input will produce two results. If S = 24,000 for example, segment overlap will be 50%, a 4.5 sec. input will produce 2 results and a 6 sec. input will produce 3 results. In all cases an additional final value is produced. That value is the mean of all segment results that are associated with segments that have speech activity factor greater than 0.45. If there are no such segments, the final result is NaN.

ctlInfo.channel = C specifies a channel number to use when the input speech is in a multi-channel .wav file. Default is C = 1.

ctlInfo.outFilename = 'myFile.txt' specifies a text file that captures WAWEnet results on a new line for each speech input processed. If the file exists it will be appended. The extension .txt will be added as needed. Default is that no .txt file is generated.

Outputs

The output for each of the N speech signals processed is in the 1 by N structure allFileInfo. The fields are:

  • allFileInfo.name, .wav filename or "MATLABVector"
  • allFileInfo.exception, description of any exception encountered
  • allFileInfo.sampleRate, native sample rate of the signal
  • allFileInfo.duration, duration of the signal in seconds
  • allFileInfo.activeLevel, active speech level of final segment in dB below overload
  • allFileInfo.activityFactor, speech activity factor of final segment
  • allfileInfo.allActivityFactors, speech activity factors for each segment
  • allfileInfo.allActiveLevels, active speech level for each segment
  • allFileInfo.netOut, output values produced by WAWEnet. Contains one value for each segment processed, plus a final value that is the mean over all segments that have speech activity factor greater than 0.45.

ctlInfo is returned as well. In addition to the five fields defined in the input section above, it also includes:

  • ctlInfo.sampleRate, native sample rate for WAWEnets, always 16,000 smp/s
  • ctlInfo.segmentLength, native segment length rate for WAWEnet, always 48,000 smp (3 s)
  • ctlInfo.outFileHeader, the header that defines results shown on screen and in any output file. This includes a starting time/date stamp.

1 Andrew A. Catellier & Stephen D. Voran, "WAWEnets: A No-Reference Convolutional Waveform-Based Approach to Estimating Narrowband and Wideband Speech Quality," ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Barcelona, Spain, 2020, pp. 331-335. ↩

2 ITU-T Recommendation P.862, "Perceptual evaluation of speech quality (PESQ)," Geneva, 2001.↩

3 ITU-T Recommendation P.863, "Perceptual objective listening quality analysis," Geneva, 2018.↩

4 R. Huber and B. Kollmeier, "PEMO-Q — A new method for objective audio quality assessment using a model of auditory perception," IEEE Trans. ASLP, vol. 14, no. 6, pp. 1902-1911, Nov. 2006.↩

5 C. H. Taal, R. C. Hendriks, R. Heusdens, and J. Jensen, "An algorithm for intelligibility prediction of time-frequency weighted noisy speech," IEEE Trans. ASLP, vol. 19, no. 7, pp. 2125-2136, Sep. 2011.↩