Dnsmos: A Non-Intrusive Perceptual Objective Speech Quality Metric to Evaluate Noise Suppressors

By Chandan K. A. Reddy, Vishak Gopal, Ross Cutler. 2020

579 citations recorded by the source corpus.

Human subjective evaluation is the "gold standard" to evaluate speech quality optimized for human perception. Perceptual objective metrics serve as a proxy for subjective scores. The conventional and widely used metrics require a reference clean speech signal, which is unavailable in real recordings. Previous no-reference approaches correlate poorly with human ratings and are not widely adopted in the research community. One of the biggest use cases of these perceptual objective metrics is to evaluate noise suppression algorithms. This paper introduces a multi-stage self-teaching based perceptual objective metric that is designed to evaluate noise suppressors. The proposed method generalizes well in challenging test conditions with a high correlation to human ratings.

Subfields: Cognitive Neuroscience, Computational Mechanics, Speech & Audio, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

In Computational Mechanics

In Speech & Audio

In Signal Processing

Citations in the corpus

VALL-E 2: Neural Codec Language Models are Human Parity Zero-Shot Text to Speech Synthesizers (2024, methodology): “Opinion Score) is used to assess the overall perceived quality of the generated speech [Reddy et al., 2021].”
SoundStorm: Efficient Parallel Audio Generation (2023, positive): “We use a MOS estimator similar to DNS-MOS (Reddy et al., 2021) to estimate the perceived audio quality of the generated samples.”
End-to-End Neural Speech Coding for Real-Time Communications (2022, neutral): “In evaluation, except the subjective listening test in Section 4.2, three metrics are used for ablation studies in Sec-Fig [20], STOI [21] and the DNSMOS [22].”
Speak, Read and Prompt: High-Fidelity Text-to-Speech with Minimal Supervision (2023, methodology): “To control the noise levels in the synthesized speech, we employ the sampling technique (Section 5) where we sample n s audio utterances corresponding to the input and select the one that has highest quality according to a no-reference audio quality model similar to DNSMOS (Reddy et al., 2021).”
FRCRN: Boosting Feature Representation Using Frequency Recurrence for Monaural Speech Enhancement (2022, unknown): “We use both DNSMOS P.835 [28] and Word Accuracy (WAcc) metrics provided by the organizer for our evaluation.”
FB-MSTCN: A Full-Band Single-Channel Speech Enhancement Method Based on Multi-Scale Temporal Convolutional Network (2022, methodology): “Perceptual evaluation of speech quality (PESQ) [16], short-time objective intelligibility (STOI) [17], source to distortion ratio (SDR) [18], and DNSMOS [19] are adopted as four evaluation metrics.”
TokenSplit: Using Discrete Speech Representations for Direct, Refined, and Transcript-Conditioned Speech Separation and Recognition (2023, methodology): “To measure signal quality, we use scale-invariant signal-to-noise ratio improvement (SI-SNRi) [30], Virtual Speech Quality Objective Listener (ViSQOL) [31], and the non-intrusive Deep Noise Suppression Mean Opinion Score (DNSMOS) [32].”
Interspeech 2021 Deep Noise Suppression Challenge (2021, methodology): “We showed that DNSMOS is more reliable than other widely used objective metrics such as PESQ, SDR, and POLQA [9].”
ICASSP 2023 Deep Noise Suppression Challenge (2023, background): “; (v) Personalized and non-personalized models for a track were sent to same subjective evaluation and ranked together, i.e. personalized and non-personalized models are treated alike and compared against each other.”
Speech Enhancement and Dereverberation With Diffusion-Based Generative Models (2022, unknown): “5) DNSMOS: The Deep Noise Suppression MOS (DNSMOS) is a reference-free metric to evaluate perceptual speech quality [60].”
Dnsmos P.835: A Non-Intrusive Perceptual Objective Speech Quality Metric to Evaluate Noise Suppressors (2021, methodology): “Recently, Deep Neural Networks (DNNs) based approaches have been proposed to estimate the speech quality scores [13, 14, 12, 15, 16, 17, 18, 19, 20].”
Torchaudio-Squim: Reference-Less Speech Quality and Intelligibility Measures in Torchaudio (2023, methodology): “Unlike more common approaches (e.g DNSMOS [9], NISQA [6]) which attempt to directly estimate MOS from a given sample, NORESQA-MOS relies on the idea of using non-matching references for a more grounded estimation [24].”
Development and Feasibility Evaluation of an Edge AI as Medical Device System for Breast Cancer Multidisciplinary Team Meetings (2026, unknown): “…with Optuna [42] (100 trials, minimising WER), and noise-adaptive inference in which Deep Noise Suppression Mean Opinion Score (DNSMOS) [43] estimated audio quality and dynamically selected default or optimised decoding parameters using thresholds between 2.0 and 5.0 (Supplementary…”