mlsanityData to 14 Oct 2026

Paradee: Distilling Kokoro-82M into an 8M-Parameter Single-Voice Text-to-Speech Model

arXiv 2610.06817 ↗First posted 5 Oct 2026cs.SD cs.AI cs.CLcited by 0

Not on the map. Why?

Abstract (arXiv)

We distill Kokoro-82M, a widely used open text-to-speech model with 54 voices, into Paradee, an 8.07M-parameter model that speaks one of them. Paradee keeps Kokoro's architecture with much narrower layers, and each of its two halves is trained separately against the frozen teacher. It has 10x fewer parameters and needs 15x less compute. We first synthesize a corpus with the teacher and keep its durations, pitch, energy and phoneme features. We then train a small text side to predict these values, and a small decoder to turn the teacher's saved values into the teacher's audio, first with spectral losses and then adversarially. Finally, we connect the two halves and quantize the weights to int8. It needs no alignment learning and no joint training, and it runs on one laptop. Stored in int8, Paradee is 8.5 MB, runs 25x faster than real time on one CPU thread, and scores 4.41 on UTMOS against the teacher's 4.52. The student initially kept a slight buzz, which we trace to the phase of voiced speech between 2 and 8 kHz. A phase-locking filter applied after synthesis removes most of it, with no training and no extra parameters. Code, model files and audio samples are at https://github.com/sahilmahendrakar/paradee

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 42 of its 51 references that are in the map. First posted on arXiv 5 Oct 2026.

Builds on

42 of 51 references.

References · 42

  1. 2015Distilling the Knowledge in a Neural Networkcited by 10,324
  2. 2016Pointer Sentinel Mixture Modelscited by 3,096
  3. 2019ALBERT: A Lite BERT for Self-supervised Learning of Language Representationscited by 2,790
  4. 2017Arbitrary Style Transfer in Real-time with Adaptive Instance Normalizationcited by 2,010
  5. 2021WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processingcited by 1,657
  6. 2020HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesiscited by 1,154
  7. 2020FastSpeech 2: Fast and High-Quality End-to-End Text to Speechcited by 717
  8. 2022High Fidelity Neural Audio Compressioncited by 687
  9. 2021Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speechcited by 561
  10. 2023High-Fidelity Audio Compression with Improved RVQGANcited by 431
  11. 2022UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022cited by 390
  12. 2019Parallel WaveGAN: A fast waveform generation model based on generative adversarial networks with multi-resolution spectrogramcited by 340
30 more
  1. 2022BigVGAN: A Universal Neural Vocoder with Large-Scale Trainingcited by 251
  2. 2017Parallel WaveNet: Fast High-Fidelity Speech Synthesiscited by 215
  3. 2023Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesiscited by 153
  4. 2023StyleTTS 2: Towards Human-Level Text-to-Speech through Style Diffusion and Adversarial Training with Large Speech Language Modelscited by 152
  5. 2024The T05 System for The VoiceMOS Challenge 2024: Transfer Learning from Deep Image Classifier to Naturalness MOS Prediction of High-Quality Synthetic Speechcited by 83
  6. 2021UnivNet: A Neural Vocoder with Multi-Resolution Spectrogram Discriminators for High-Fidelity Waveform Generationcited by 74
  7. 2022StyleTTS: A Style-Based Generative Model for Natural and Diverse Text-to-Speech Synthesiscited by 33
  8. 2022iSTFTNet: Fast and Lightweight Mel-Spectrogram Vocoder Incorporating Inverse Short-Time Fourier Transformcited by 32
  9. 2021LightSpeech: Lightweight and Fast Text to Speech with Neural Architecture Searchcited by 28
  10. 2023Phoneme-Level BERT for Enhanced Prosody of Text-to-Speech with Grapheme Predictionscited by 23
  11. 2018Neural source-filter-based waveform model for statistical parametric speech synthesiscited by 22
  12. 2019Neural Harmonic-plus-Noise Waveform Model with Trainable Maximum Voice Frequency for Text-to-Speech Synthesiscited by 11
  13. 2022Neural Speech Phase Prediction based on Parallel Estimation Architecture and Anti-Wrapping Lossescited by 11
  14. 2023HiFTNet: A Fast High-Quality Neural Vocoder with Harmonic-plus-Noise Filter and Inverse Short Time Fourier Transformcited by 11
  15. 2022Lightweight and High-Fidelity End-to-End Text-to-Speech with Multi-Band Generation and Inverse Short-Time Fourier Transformcited by 9
  16. 2024StyleTTS-ZS: Efficient High-Quality Zero-Shot Text-to-Speech Synthesis with Distilled Time-Varying Style Diffusioncited by 7
  17. 2024FreeV: Free Lunch For Vocoders Through Pseudo Inversed Mel Filtercited by 6
  18. 2023APNet2: High-quality and High-efficiency Neural Vocoder with Direct Prediction of Amplitude and Phase Spectracited by 5
  19. 2019FastSpeech: Fast, Robust and Controllable Text to Speechcited by 4
  20. 2022Nix-TTS: Lightweight and End-to-End Text-to-Speech via Module-wise Distillationcited by 4
  21. 2023APNet: An All-Frame-Level Neural Vocoder Incorporating Direct Prediction of Amplitude and Phase Spectracited by 4
  22. 2023iSTFTNet2: Faster and More Lightweight iSTFT-Based Neural Vocoder Using 1D-2D CNNcited by 4
  23. 2021On the Interplay Between Sparsity, Naturalness, Intelligibility, and Prosody in Speech Synthesiscited by 3
  24. 2024Wavehax: Aliasing-Free Neural Waveform Synthesis Based on 2D Convolution and Harmonic Prior for Reliable Complex Spectrogram Estimationcited by 2
  25. 2025SupertonicTTS: Towards Highly Efficient and Streamlined Text-to-Speech Systemcited by 2
  26. 2026Attacking UTMOS: Probing the Robustness of a Speech Quality Assessment Modelcited by 2
  27. 2026The Limits of Reference-Free Speech Quality Metrics as Evaluators and Rewards on Modern Text-to-Speechcited by 2
  28. 2025A Distilled Low-Latency Neural Vocoder with Explicit Amplitude and Phase Predictioncited by 1
  29. 2026sanoTTS: The Smallest Real-Time Neural TTS on a General-Purpose Microcontrollercited by 1
  30. 2026Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speechcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Training a Compact, Strong, and Realistic Text-to-Speech Model
  2. 2026Raon-OpenTTS: Open Models and Data for Robust Text-to-Speech ↗
  3. 2026CuteTTS: Efficient and High-Quality Speech Synthesis via Autoregressive Modeling of Continuous Latents
  4. 2026DriftTTS: Few-Step Text-to-Speech Without Distillation via Distribution-Matching Drift
  5. 2026Stride-k Subsampling: Train-Free Audio Token Reduction for Whisper
  6. 2025TESU-LLM: Training Speech-LLMs Without Speech via Unified Encoder Alignment
  7. 2023Voicebox: Text-Guided Multilingual Universal Speech Generation at Scale
  8. 2026GEPARD - Generative, Prosody-aware, Autoregressive text-to-speech model for Realtime Dialogue