PESQ and STOI Speech Quality Metrics
PESQ and STOI are standard objective metrics that score how good processed speech sounds and how understandable it is, without needing human listeners.
Overview
They let engineers benchmark codecs, noise reducers, and speech-enhancement models automatically.
Deep Dive
PESQ (Perceptual Evaluation of Speech Quality), standardized as ITU-T P.862, predicts the perceived quality of speech, mainly for telephone and codec testing. It compares a clean reference signal to a degraded one and outputs a score on a MOS-like scale (roughly -0.5 to 4.5), modeling human auditory perception. STOI (Short-Time Objective Intelligibility), introduced in 2010, instead predicts intelligibility: how many words a listener would actually understand. It correlates short-time temporal envelopes of clean and processed speech across frequency bands, producing a score from 0 to 1. Both are intrusive (reference-based) metrics. PESQ answers 'does it sound good?' while STOI answers 'can you understand it?' Together they are the default evaluation tools for speech enhancement, denoising, and dereverberation systems.
Technical Insight
Both metrics are intrusive: they align a clean reference with the degraded signal before scoring. PESQ maps both signals onto a psychoacoustic loudness scale (Bark bands), computes perceptual disturbance over time, and regresses it to a MOS-like value. STOI splits speech into one-third-octave bands, takes short ~400 ms envelope segments, clips and normalizes them, then computes the correlation between reference and degraded envelopes. Averaging those correlations yields the 0-to-1 intelligibility score.
Strategic Impact
Access and reach
It improves accessibility through transcription, narration, and voice interfaces.
Cost and budget
Media teams can ship polished audio faster with smaller budgets.
Speed and scale
Customer-facing systems can process spoken interactions at larger scale.
The Future of PESQ and STOI Speech Quality Metrics
Because PESQ and STOI need a clean reference, research is shifting toward non-intrusive, reference-free metrics like DNSMOS and NISQA that score quality from the degraded signal alone using neural networks. Newer deep-learning models are also trained to predict human MOS directly. Still, PESQ and STOI remain entrenched benchmarks, and a key trend is making them differentiable so they can be used directly as training loss functions for speech-enhancement networks rather than only as after-the-fact evaluations.
Real-World Implementation
Benchmarking speech-enhancement and noise-suppression models on standard test sets
Comparing telephone and VoIP codec quality during network engineering
Tuning hearing-aid and cochlear-implant processing for maximum intelligibility
Validating dereverberation algorithms in conferencing and voice-assistant pipelines
Risks & Guardrails
Voice misuse and impersonation risks increase when consent is missing.
Accuracy can drop across accents, dialects, or noisy environments.
Synthetic audio can be mistaken for authentic speech without clear labeling.
Implementation Roadmap
Obtain explicit consent for voice capture, cloning, and reuse.
Test quality across diverse speakers and background conditions.
Define when a human must review or approve outputs.
Label synthetic audio and keep provenance records for accountability.
Keep Exploring
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the PESQ and STOI Speech Quality Metrics quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Next guide
Speech Synthesis Quality
Frequently asked questions
What is PESQ and STOI Speech Quality Metrics?
PESQ and STOI are standard objective metrics that score how good processed speech sounds and how understandable it is, without needing human listeners. They let engineers benchmark codecs, noise reducers, and speech-enhancement models automatically.
What does PESQ primarily measure?
PESQ (ITU-T P.862) predicts perceived speech quality on a MOS-like scale.
What does STOI primarily predict?
STOI estimates intelligibility, i.e. how understandable the speech is, on a 0-to-1 scale.
Both PESQ and STOI are described as 'intrusive' metrics. What does that mean?
Intrusive metrics compare the degraded signal against a clean reference to compute a score.
What is the approximate score range of STOI?
STOI outputs a value between 0 and 1, where higher means more intelligible.
PESQ models human hearing using which kind of perceptual frequency scale?
PESQ maps signals onto a psychoacoustic loudness representation using Bark-band processing.