Supplementary material for the INTERSPEECH 2023 paper Why We Should Report the Details in Subjective Evaluation of TTS More Rigorously
The following table is the 80 paper we survey; all of them are from INTERSPEECH 2022.
In the table, the v
means that the paper reports the factor, and x
means that the paper do not report the factor.
- The column # of raters and rated items considers four factors: (1) the total number of unique raters, (2) the number of samples a rater rates, (3) how many raters evaluate a single sample, and (4) how many audio samples in total.
If the paper report none of the four factors, we put a
x
; if the paper report all four factors, we put av
; if the paper report some but not all the four factors, we put a△
. - For the column Instruction, we check if the paper provides either the rating scales or the questions presented to the evaluators.
- For the column Increment, we check if the paper uses a 1-point increment or 0.5-point increment in the MOS tests.
For those we cannot know what increment the paper uses, we put a
x
. Some paper just say "from 1~5", but this description is too vague for us to know if it is an 1-increment or a 0.5-increment, so we still put ax
in this case. If the paper explicitly cites P.800 or P.808 and says the protocol is followed in the experiment but does not specify the increment of MOS scores, we treat it as 1-increment since 1-increment is used in those protocols.
If you find the survey results to be wrong or you have questions on the survey results, please contact Cheng-Han Chiang via [email protected].
Title | Platform | Qualification | Post-filter | Language background | Location | # of raters and rated items | Payment | Instructions | Increment (MOS) |
---|---|---|---|---|---|---|---|---|---|
SANE-TTS: Stable And Natural End-to-End Multilingual Text-to-Speech | Mturk | x | x | v | x | △ | x | x | 1 |
Enhancement of Pitch Controllability using Timbre-Preserving Pitch Augmentation in FastPitch | x | x | x | v | x | △ | x | v | 1 |
Speaking Rate Control of end-to-end TTS Models by Direct Manipulation of the Encoder’s Output Embeddings | Prolific | x | x | x | x | v | x | v | - |
TriniTTS: Pitch-controllable End-to-end TTS without External Aligner | Mturk | x | x | x | x | △ | x | x | 1 |
JETS: Jointly Training FastSpeech2 and HiFi-GAN for End to End Text to Speech | Mturk | x | x | x | v | △ | x | v | 1 |
EdiTTS: Score-based Editing for Controllable Text-to-Speech | Mturk | x | x | x | x | △ | x | x | 1 |
SpeechPainter: Text-conditioned Speech Inpainting Zalan Borsos | x | x | x | x | x | △ | x | v | 1 |
Neural Lexicon Reader: Reduce Pronunciation Errors in End-to-end TTS by Leveraging External Textual Knowledge | x | x | x | x | x | △ | x | x | x |
DocLayoutTTS: Dataset and Baselines for Layout-informed Document-level Neural Speech Synthesis | x | v | x | v | x | △ | x | x | 0.5 |
Mixed-Phoneme BERT: Improving BERT with Mixed Phoneme and Sup-Phoneme Representations for Text to Speech | x | x | x | v | x | △ | x | x | 1 |
An Efficient and High Fidelity Vietnamese Streaming End-to-End Speech Synthesis | x | x | x | v | v | △ | x | x | x |
Environment Aware Text-to-Speech Synthesis | x | x | x | x | x | △ | x | x | x |
A Study of Modeling Rising Intonation in Cantonese Neural Speech Synthesis | x | x | x | v | x | △ | x | x | x |
Transfer Learning Framework for Low-Resource Text-to-Speech using a Large-Scale Unlabeled Speech | x | x | x | x | x | △ | x | x | 1 |
DRSpeech: Degradation-Robust Text-to-Speech Synthesis with Frame-Level and Utterance-Level Acoustic Representation Learning | x | x | x | x | x | △ | x | x | x |
MSR-NV: Neural Vocoder Using Multiple Sampling Rates | x | v | x | x | x | △ | x | x | - |
SpecGrad: Diffusion Probabilistic Model based Neural Vocoder with Adaptive Noise Spectral Shaping | x | x | x | v | v | △ | v | v | 0.5 |
Bunched LPCNet2: Efficient Neural Vocoders Covering Devices from Cloud to Edge | Mturk | x | x | x | x | △ | x | v | 1 |
Hierarchical and Multi-Scale Variational Autoencoder for Diverse and Natural Non-Autoregressive Text-to-Speech | x | x | x | v | x | △ | x | x | x |
End-to-end LPCNet: A Neural Vocoder With Fully-Differentiable LPC Estimation' | x | x | x | x | x | △ | x | v | 1 |
WavThruVec: Latent speech representation as intermediate features for neural speech synthesis | Mturk | v | x | x | v | △ | x | x | 1 |
Fast Grad-TTS: Towards Efficient Diffusion-Based Speech Generation on CPU | Mturk | v | x | x | x | △ | x | x | 0.5 |
Simple and Effective Unsupervised Speech Synthesis | Mturk | x | x | x | x | △ | x | x | 1 |
Unified Source-Filter GAN with Harmonic-plus-Noise Source Excitation Generation | x | x | x | x | x | △ | x | x | x |
Speaker Attractor Text to Speech, Learning to Speak by Learning to Separate | x | x | x | x | x | △ | x | v | 0.5 |
A Multi-Scale Time-Frequency Spectrogram Discriminator for GAN-based Non-Autoregressive TTS | Mturk | x | x | x | x | △ | x | x | x |
RetrieverTTS: Modeling Decomposed Factors for Text-Based Speech Insertion | x | x | x | x | x | △ | x | x | x |
FlowVocoder: A small Footprint Neural Vocoder based Normalizing Flow for Speech Synthesis | x | x | x | x | x | △ | x | x | 1 |
DelightfulTTS 2: End-to-End Speech Synthesis with Adversarial Vector-Quantized Auto-Encoders | x | x | v | x | x | △ | x | x | 0.5 |
AdaVocoder: Adaptive Vocoder for Custom Voice | x | x | x | x | x | △ | x | x | x |
RefineGAN: Universally Generating Waveform Better than Ground Truth with Highly Accurate Pitch and Intensity Responses | x | x | x | x | x | △ | x | x | x |
VQTTS: High-Fidelity Text-to-Speech Synthesis with Self-Supervised VQ Acoustic Feature | x | x | x | x | x | △ | x | x | x |
IMPROVING GAN-BASED VOCODER FOR FAST AND HIGH-QUALITY SPEECH SYNTHESIS | x | x | x | v | x | △ | x | v | 1 |
SoftSpeech: Unsupervised Duration Model in FastSpeech 2 | x | x | x | x | x | △ | x | v | 1 |
A Multi-Stage Multi-Codebook VQ-VAE Approach to High-Performance Neural TTS | x | x | x | x | x | △ | x | x | 0.5 |
Joint Modeling of Multi-Sample and Subband Signals for Fast Neural Vocoding on CPU | x | x | x | x | x | △ | x | v | 1 |
MISRNet: Lightweight Neural Vocoder Using Multi-Input Single Shared Residual Blocks | x | x | x | x | x | △ | x | v | 1 |
A compact transformer-based GAN vocoder | x | x | x | x | x | △ | x | x | x |
Diffusion Generative Vocoder for Fullband Speech Synthesis Based on Weak Third-order SDE Solver | x | x | x | x | x | x | v | x | x |
From Start to Finish: Latency Reduction Strategies for Incremental Speech Synthesis in Simultaneous Speech-to-Speech Translation | Mturk | x | x | x | x | △ | x | x | x |
Unified Accent Estimation Method Based on Multi-Task Learning for Japanese Text-to-Speech | x | x | x | v | x | △ | x | v | 1 |
Vocal effort modeling in neural TTS for improving the intelligibility of synthetic speech in noise | x | x | x | v | v | △ | x | x | x |
TTS-by-TTS 2: Data-selective augmentation for neural speech synthesis using ranking support vector machine with variational autoencoder | x | x | x | v | v | △ | x | x | x |
Glow-WaveGAN 2: High-quality Zero-shot Text-to-speech Synthesis and Any-to-any Voice Conversion | x | x | x | x | x | △ | x | x | x |
Content-Dependent Fine-Grained Speaker Embedding for Zero-Shot Speaker Adaptation in Text-to-Speech Synthesis | x | x | x | v | x | △ | x | x | 1 |
Acoustic Modeling for End-to-End Empathetic Dialogue Speech Synthesis Using Linguistic and Prosodic Contexts of Dialogue History | x | x | x | x | x | △ | x | v | x |
Emphasis Control for Parallel Neural TTS | x | x | x | v | v | △ | x | v | 1 |
Combining conversational speech with read speech to improve prosody in Text-to-Speech synthesis | Prolific | v | v | v | v | △ | x | v | 1 |
Exploring Timbre Disentanglement in Non-Autoregressive Cross-Lingual Text-to-Speech | x | x | x | v | x | △ | x | x | 0.5 |
WeSinger: Data-augmented Singing Voice Synthesis with Auxiliary Losses | x | x | x | x | x | △ | x | v | x |
Decoupled Pronunciation and Prosody Modeling in Meta-Learning-based Multilingual Speech Synthesis | Mturk | x | x | v | x | △ | x | v | x |
KaraTuner: Towards End-to-End Natural Pitch Correction for Singing Voice in Karaoke | x | v | x | x | x | △ | x | x | x |
Learn2Sing 2.0: Diffusion and Mutual Information-Based Target Speaker SVS by Learning from Singing Teacher | x | x | x | x | x | x | x | x | x |
SingAug: Data Augmentation for Singing Voice Synthesis with Cycle-consistent Training Strategy | x | v | x | x | x | △ | x | v | x |
Muskits: an End-to-end Music Processing Toolkit for Singing Voice Synthesis | x | v | x | x | x | △ | x | v | x |
Pronunciation Dictionary-Free Multilingual Speech Synthesis by Combining Unsupervised and Supervised Phonetic Representations | x | x | x | v | x | △ | x | x | 0.5 |
Towards High-fidelity Singing Voice Conversion with Acoustic Reference and Contrastive Predictive Coding | x | v | x | x | x | △ | x | v | x |
Towards Improving the Expressiveness of Singing Voice Synthesis with BERT Derived Semantic Information | x | x | x | v | x | △ | x | x | 1 |
Normalization of code-switched text for speech synthesis | x | v | x | v | x | △ | x | x | x |
Synthesizing Near Native-accented Speech for a Non-native Speaker by Imitating the Pronunciation and Prosody of a Native Speaker | Mturk | x | x | v | x | △ | x | x | x |
A Hierarchical Speaker Representation Framework for One-shot Singing Voice Conversion | x | x | x | v | x | △ | x | x | x |
Predicting VQVAE-based Character Acting Style from Quotation-Annotated Text for Audiobook Speech Synthesis | x | x | x | x | x | △ | x | x | x |
Adversarial and Sequential Training for Cross-lingual Prosody Transfer TTS | x | x | x | x | x | x | x | x | x |
FluentTTS: Text-dependent Fine-grained Style Control for Multi-style TTS | x | x | x | x | x | △ | x | v | x |
Few Shot Cross-Lingual TTS Using Transferable Phoneme Embedding | x | x | x | x | x | △ | x | x | x |
Training Text-To-Speech Systems From Synthetic Data: A Practical Approach For Accent Transfer Tasks | x | x | x | x | x | x | x | v | x |
Spoken-Text-Style Transfer with Conditional Variational Autoencoder and Content Word Storage | x | x | x | v | x | △ | x | x | x |
Analysis of expressivity transfer in non-autoregressive end-to-end multispeaker TTS systems | x | x | x | x | v | △ | x | v | x |
Daft-Exprt: Cross-Speaker Prosody Transfer on Any Text for Expressive Speech Synthesis | x | x | v | v | x | △ | x | x | x |
Language Model-Based Emotion Prediction Methods for Emotional Speech Synthesis Systems | x | x | x | v | x | △ | x | v | x |
Text aware Emotional Text-to-speech with BERT | x | x | x | v | x | △ | x | v | x |
Transplantation of Conversational Speaking Style with Interjections in Sequence-to-Sequence Speech Synthesis | x | v | x | x | x | x | x | x | x |
Cross-speaker Emotion Transfer Based On Prosody Compensation for End-to-End Speech Synthesis | x | x | x | v | x | △ | x | v | x |
Self-supervised Context-aware Style Representation for Expressive Speech Synthesis | UHRS | x | x | x | x | △ | x | v | x |
Automatic Prosody Annotation with Pre-Trained Text-Speech Model | x | x | x | x | x | △ | x | v | x |
Enhancing Word-Level Semantic Representation via Dependency Structure for Expressive Text-to-Speech Synthesis | x | v | x | x | x | △ | x | x | x |
Towards Multi-Scale Speaking Style Modelling with Hierarchical Context Information for Mandarin Speech Synthesis | x | x | x | v | x | △ | x | x | 1 |
Towards Cross-speaker Reading Style Transfer on Audiobook Dataset | x | x | x | v | x | △ | x | x | x |
CALM: Constrastive Cross-modal Speaking Style Modeling for Expressive Text-to-Speech Synthesis | x | x | x | v | x | △ | x | x | 1 |
Improve emotional speech synthesis quality by learning explicit and implicit representations with semi-supervised training | x | x | x | v | x | △ | x | x | 0.5 |
Here is one of the surveys that is used when conducting MOS tests with the students from our college: Survey example. The link and the survey do not contain any information about the authors.
Here is the interface on Mturk that is used for all the experiments except recruiting raters from our college.