Issues

• [License clarification] Can Qwen3-TTS VoiceDesign outputs be used as training data for private single-speaker SFT?

  QwenLM/Qwen3-TTS#375 · Filter by author gpsoulmoon-sys opened on Sep 30, 2026

• finetuning/sft_12hz.py hard-codes attn_implementation="flash_attention_2" — official fine-tuning fails out-of-the-box on ROCm (no flash-attn build)

  QwenLM/Qwen3-TTS#372 · Filter by author AIwork4me opened on Sep 20, 2026

• Finetuning loss misaligned with generation (double shift + leaked sub-talker hidden) — please merge #278

  QwenLM/Qwen3-TTS#371 · Filter by author thewh1teagle opened on Sep 16, 2026

• 25Hz tokenizer: qkv_attention_manual never masks padded keys (bool masked_fill), short attention windows get wrong outputs

  QwenLM/Qwen3-TTS#369 · Filter by author raghav-rathi opened on Sep 13, 2026

• finetuning/sft_12hz.py: checkpoint save is non-atomic (can silently leave base weights) and peaks at 2x model size in host RAM

  QwenLM/Qwen3-TTS#365 · Filter by author Stylegunhawk opened on Sep 3, 2026
  1

• Feature request: native Italian preset voices (male + female)

  inactive
  QwenLM/Qwen3-TTS#359 · Filter by author solyarisoftware opened on Aug 19, 2026
  2

• [Commercial use clarification] May outputs from preset voice Sohee be used in monetized YouTube videos?

  inactive
  QwenLM/Qwen3-TTS#353 · Filter by author lovisjin opened on Aug 2, 2026
  1

• Multi-concurrency inference not supported

  inactive
  QwenLM/Qwen3-TTS#350 · Filter by author supy0001 opened on Jul 24, 2026
  1

• Verify evals on Papers with Code

  inactive
  QwenLM/Qwen3-TTS#347 · Filter by author NielsRogge opened on Jul 22, 2026
  1

• Fine-tuned 1.7B voice: short/opening utterances unstable (wrong timbre, occasional gender flip) while long-form is stable

  inactive
  QwenLM/Qwen3-TTS#343 · Filter by author AgentOffersEthan opened on Jul 10, 2026
  2

• Voice-clone (ICL) generation echoes the tail of the reference audio before the requested text

  inactive
  QwenLM/Qwen3-TTS#341 · Filter by author karlanghas opened on Jul 9, 2026
  2

• The talker_hidden_states not obtained correctly

  inactive
  QwenLM/Qwen3-TTS#337 · Filter by author howitry opened on Jul 6, 2026
  9

• Any plan to release 25Hz tts model ？

  inactive
  QwenLM/Qwen3-TTS#294 · Filter by author xiaoyangnihao opened on Apr 12, 2026
  6

• Finetuning Base results in progressively faster speech with every epoch

  inactive
  QwenLM/Qwen3-TTS#179 · Filter by author dragonfyre13 opened on Feb 5, 2026
  37

• The audio generated for Japanese and Korean is incomplete.

  inactive
  QwenLM/Qwen3-TTS#55 · Filter by author miles-god opened on Jan 24, 2026
  10