Speech Quality Assessment

By P. Loizou. 2011

195 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, 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 Artificial Intelligence

1. Xu Tan
2. Tao Qin
3. Tie-Yan Liu
4. Sheng Zhao
5. Frank K. Soong
6. Yi Ren
7. Zhou Zhao
8. Lei He

In Computer Vision and Pattern Recognition

1. Tao Qin
2. Tie-Yan Liu
3. Xu Tan
4. Zhou Zhao
5. Enhong Chen
6. Renqian Luo
7. Rui Wang
8. G. Wang

In Signal Processing

1. Xu Tan
2. Tao Qin
3. Tie-Yan Liu
4. Yi Ren
5. Sheng Zhao
6. Zhou Zhao
7. Frank K. Soong
8. Lei He

Citations in the corpus

• A Wavenet for Speech Denoising (2017, background): “Participants were able to give a score between 1–5, with a 1 being described as “degraded speech with very intrusive background" and a 5 being “not degraded speech with unnoticeable background" [10, 16].”
• FastSpeech 2: Fast and High-Quality End-to-End Text to Speech (2020, methodology): “We find that removing the energy variance (Row 2 in both subtables) in FastSpeech 2 and 2s results in performance drop in terms of voice quality (-0.045 and -0.150 CMOS respectively), indicating that energy variance can slightly improve the voice quality for FastSpeech 2, but more effective for Fast”
• Composition of Deep and Spiking Neural Networks for Very Low Bit Rate Speech Coding (2016, methodology): “For this qualitative comparison, we employed a 5-point scale ABX subjective evaluation listening test [69].”
• DCTTS: Discrete Diffusion Model with Contrastive Learning for Text-to-Speech Generation (2023, methodology): “3, by comparing the CMOS[21] of DCTTS(w/o TCLL) and DCTTS, it is shown that the TCLL has a positive effect on the quality of generated speech.”
• Mixed-Phoneme BERT: Improving BERT with Mixed Phoneme and Sup-Phoneme Representations for Text to Speech (2022, methodology): “Total 60 text sentences are selected for the subjective test, with at least 15 native judges for the five-score MOS test and 12 judges for Comparison MOS (CMOS) [29] for each test case.”
• A DNN-based Mandarin-Tibetan cross-lingual speech synthesis (2018, methodology): “At the same time, the maximum likelihood parameter generation (MLPG) algorithm [23] is used to generate the speech parameters from acoustic models.”
• Revisiting Over-Smoothness in Text to Speech (2022, methodology): “Subjective Evaluation To evaluate the perceptual quality, we conduct the MOS and CMOS (Loizou, 2011) tests.”
• Lightspeech: Lightweight and Fast Text to Speech with Neural Architecture Search (2021, methodology): “Audio Quality To evaluate the quality of synthesized speech, we perform CMOS [25] evaluation on the test set.”
• Reasons why current speech-enhancement algorithms do not improve speech intelligibility and suggested solutions (2011, methodology): “The time-domain segmental (and overall) SNR measure has been used widely and frequently for evaluating speech quality in speech coding and enhancement applications [10], [11].”
• Prompttts: Controllable Text-To-Speech With Text Descriptions (2022, methodology): “We also conduct comparison MOS (CMOS) [21] to compare the speech quality between PromptTTS and the two-stage baseline system.”
• An Overview of Voice Conversion and Its Challenges: From Statistical Modeling to Deep Learning (2020, background): “Mel-cepstral distortion (MCD) [246] for spectrum, and PCC [247] and RMSE [248]–[250] for prosody.”
• s-Transformer: Segment-Transformer for Robust Neural Speech Synthesis (2020, background): “We )[23] and 15 judges for MOS test for each test case.”
• Torchaudio-Squim: Reference-Less Speech Quality and Intelligibility Measures in Torchaudio (2023, background): “Note that, there are certain objective metrics which may not require reference clean speech [1, 2], but for the purposes of this paper objective metrics would refer to the more common cases where reference clean speech is required [1].”
• Hierarchical Prosody Modeling for Non-Autoregressive Speech Synthesis (2020, methodology): “R, P+N) are even comparable to the samples reconstructed from ground-truth mel-spectrograms, reflecting the effectiveness of prosody modeling. 4.4.2. Pairwise subjective scores We further conduct CMOS [29] and AXY test [1] to compare hierarchical models and non-hierarchical models pairwisely. vanilla”