proomt

Search

Search posts, papers, and topics

All posts

Hugging Face Daily PapersKunat Pipatanakul, Potsawee Manakul, Warit Sirichotedumrong1 min readpaperadvanced

Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech

Read on arxiv.org

Summary

Researchers train a compact 82 M‑parameter Thai fixed‑voice TTS model using synthetic speech generated by a large voice‑cloning teacher, requiring only a 15‑second real reference. The student achieves 68.2% keyword accuracy and 91.4% pause precision, outperforming its teacher on pause placement and enabling on‑device inference.

  • Synthetic data from a large voice‑cloning model lets a 15 s real audio reference produce an 82 M‑parameter fixed‑voice Thai TTS model.
  • Text preprocessing, generation filtering, and rejection sampling are essential to mitigate teacher errors while preserving coverage.
  • The student reaches 68.2% challenge‑set keyword accuracy (85.5% of Gemini 3.1) and 91.4% pause precision, surpassing its teacher on pause placement.
  • Character error rates are 3.7% for Thai and 1.1% for English loanwords, handling tonal and code‑switching challenges.

Teams building on‑device TTS for low‑resource languages should care because this shows a practical way to get a compact, high‑quality fixed‑voice model from minimal real data.

7/10
ttssynthetic datathaispeech synthesislow resource

Related reading

  2. VākQA: A Benchmark and Evaluation Study for Telugu Spoken Factoid Question Answering

    The authors present VākQA, a 2,001‑question spoken factoid QA benchmark for Telugu with audio, transcriptions, and human‑verified answers, and they validate automatic evaluation methods against human ratings. Using this setup they show that translation loses cultural nuance, ASR errors alter meaning, and cascaded ASR‑MT errors degrade model performance.

    Hugging Face Daily Papersarxiv.org1 minpaper

  3. How Far Can Synthetic Data Take Thai OCR?

    The paper dissects which aspects of synthetic Thai OCR data (typeface diversity, 2‑D layout, real handwriting glyphs) actually help transfer to real documents. Using those insights they train Wayu-Paxa-OCR-Zero without any real OCR labels and achieve character error rates comparable to or better than supervised baselines.

    Hugging Face Daily Papersarxiv.org1 minpaper

  4. StepAudio 3 Realtime Technical Report

    StepAudio 3 Realtime is an audio‑language foundation model that runs a continuous listen‑converse‑think‑act loop. It introduces Deep Perception for rich acoustic cue extraction, Seamless Duplex for handling pauses/back‑channels, and a Think‑While‑Speaking mechanism that lets the model reason in parallel with speech output. On benchmarks it scores 73.0 macro avg on StepAudioChat, 90.6 on MMSU, 98.…

    Hugging Face Daily Papersarxiv.org2 minpaper

  5. OmniVChat: Synthesizing, Benchmarking, and Training for Native Audio-Visual Dialogue

    OmniVChat defines native audio‑visual dialogue where a model consumes raw audio and video streams and replies in text. The authors build OmniVChat‑Studio, a multi‑agent simulator that generates single‑ and multi‑turn audio‑visual conversations, and use it to create OmniVChat‑Bench, a benchmark covering five dialogue abilities. They also propose OmniVChat‑RL, a reinforcement‑learning reward that b…

    Hugging Face Daily Papersarxiv.org2 minpaper

  6. StepAudio 3 Gen Technical Report

    StepAudio 3 Gen is a general‑purpose audio generation model that replaces diffusion with a discrete autoregressive generator over residual vector quantization tokens. Using a 16‑layer RVQ tokenizer and progressive pretraining, it reaches state‑of‑the‑art zero‑shot TTS and voice‑design performance while handling speech, vocals, sound effects, and music.

    Hugging Face Daily Papersarxiv.org2 minpaperHN2