ApproximatePerceptualSpeechQuality<T>
Models & Types · AiDotNet.Metrics
A PESQ-like speech quality approximation. This is NOT ITU-T P.862 (PESQ) and does not reproduce its scores.
How It Works
This metric blends STOI and segmental SNR into a number on the MOS-LQO range (-0.5 to 4.5) using an
empirical mapping. It does not implement the ITU-T P.862 / P.862.2 perceptual model (level and IRS
filtering, time alignment, auditory transform, disturbance aggregation), and its values do not match
the official reference implementation or the python pesq package. Do not report its output as
PESQ or compare it with published PESQ results.
Use it only as a rough relative quality indicator; higher values indicate better quality.
Constructors
| Constructor | Summary |
|---|---|
ApproximatePerceptualSpeechQuality(int sampleRate) |
Initializes a new instance of the PESQ-like approximation (not ITU-T P.862). |
Methods
| Method | Summary |
|---|---|
T Compute(Tensor<T> degraded, Tensor<T> reference) |
Computes the PESQ-like approximation between degraded and reference speech (not an ITU-T P.862 score). |