package Visqol

Get desktop application:
View/edit binary Protocol Buffers messages

• double moslqo = 1

  The "Mean Opinion Score - Listening Quality Objective" for the degraded signal, following the comparison to the reference signal. Produced using an SVR model.

• double vnsim = 2

  Mean of FVNSIM

• repeated double fvnsim = 3

  Mean of per patch mean frequency band similarity across all patches. The order of the elements is lowest to highest in frequency bands.

• repeated double fvnsim10 = 11

  Mean of 10th percentile frequency band similarity across all patches.

• repeated double fstdnsim = 8

  Stddev of per patch mean frequency band similarity across all patches. The order of the elements is lowest to highest in frequency bands.

• repeated double fvdegenergy = 9

  Mean/Average energy over patches for each frequency band in the degraded signal.

• repeated double center_freq_bands = 4

  Lowest to highest center frequency bands:

• repeated SimilarityResultMsg.PatchSimilarityMsg patch_sims = 5

  The patch comparison results for each patch compared between the reference and degraded signals.

• string reference_filepath = 6

  If ViSQOl was used at the command line to process a reference and degraded filepath pair for comparison, this will hold the reference filepath.

• string degraded_filepath = 7

  If ViSQOl was used at the command line to process a reference and degraded filepath pair for comparison, this will hold the degraded filepath.

• double alignment_lag_s = 10

  If the degraded audio was additionally aligned, this will store the value applied, in seconds. If it is positive, then the degraded audio was later than the reference, if it is negative then the reference audio was later than the degraded. To align audio, apply the lag to the audio file that was later by prepending 0 or moving indices.

Contains info related to the similarity result for each patch.

Used in: SimilarityResultMsg

• double similarity = 1

  Similarity score for this patch.

• repeated double freq_band_means = 2

  Stored lowest to highest.

• double ref_patch_start_time = 3

  The time (in sec) where this patch starts in the reference signal.

• double ref_patch_end_time = 4

  The time (in sec) where this patch end in the reference signal.

• double deg_patch_start_time = 5

  The time (in sec) where this patch starts in the degraded signal.

• double deg_patch_end_time = 6

  The time (in sec) where this patch ends in the degraded signal.

• optional VisqolConfig.VisqolAudioInfo audio = 1

• optional VisqolConfig.VisqolOptions options = 2

Contains info related to the input signals.

Used in: VisqolConfig

• int32 sample_rate = 1

  The sample rate of the input signals. Both signals must have the same sample rate. This value must be supplied.

Contains config info for ViSQOL

Used in: VisqolConfig

• bool output_mos_score = 1

  Not yet supported.

• string svr_model_path = 2

  The path to a svr model file. If not supplied, the default model is used.

• bool use_speech_scoring = 3

  If true, the input audio files will be compared using the ViSQOL speech mode. See use_unscaled_speech_mos_mapping below for details regarding the mapping of NSIM to MOS-LQO.

• bool detect_voice_activity = 4

  Not yet supported.

• bool allow_unsupported_sample_rates = 5

  Currently, only 48k samples rates are supported by ViSQOL Audio - (Speech mode does not have this restriction). To override this and run with non-48k input, set this bool to true.

• bool use_unscaled_speech_mos_mapping = 6

  When using the speech comparison mode, a value of false for this bool will cause a perfect NSIM score of 1.0 to be mapped to a perfect MOS-LQO of 5.0. If this bool is instead set to true, a perfect NSIM score will instead be mapped to a MOS-LQO of ~4.x.

• int32 search_window_radius = 7

  If provided, this value will be used to determine how far the comparison algorithm will search to find the most optimal match for a given reference frame.

• bool use_lattice_model = 8

  If true, use a deep lattice model to map similarity to MOS. This generally provides more accurate results than the default polynomial or SVR. This is recommended unless comparing to historic conformance scores. The binary default for this is `true`.