Repository Issue Activity (beta)

#780 feat(stt): SRT/VTT cues are one-per-chunk and too long for subtitles

#813 MLX Kokoro TTS produces NaN / silent audio on Apple Silicon — all voices, all languages

#815 Kokoro TTS produces NaN + astronomically large audio on all languages (MLX numerical instability)

#818 Server: all Qwen3-TTS models fail to load — 'str' object has no attribute '__module__' (reproduced in 3 clean environments)

#782 Qwen3-ASR ignores --max-parallel-segments; decodes chunks sequentially

#816 Long audio (>5min) causes Metal OOM in Nemotron ASR — log_mel_spectrogram and create_chunked_limited_mask process full input at once

#796 Add Confucius4-TTS: multilingual zero-shot voice-cloning TTS (MLX port available)

#809 Feature Request: Add support for higgs-audio-v3 STT from Boson AI

#784 Kokoro generate() fails with [broadcast_shapes] Shapes (1,N,1) and (1,N+300,9) on 0.4.4 (works on 0.4.1)

#786 Bug: Kokoro audio generation crashes with ValueError (broadcast shapes mismatch) in v0.4.4

#803 Kokoro SineGen._f02sine is not length-preserving → broadcast_shapes failure on most utterances

#714 OmniVoice generates empty WAVs when audio tokenizer load fails (which is always)

#787 Bug: Qwen3 TTS cloning crash

#794 crash on TTS on every model

#789 SimpleChat `.xcodeproj` folder/bundle is missing

#1 TTS and STS Models to port to MLX-Audio (Roadmap)

#761 Feature: Add FunASR/SenseVoice models for STT on Apple Silicon

#758 Feature Request: Add SenseVoice/Paraformer models for STT

#756 feat: Add SenseVoice model support for ASR

#755 Feature Request: Add SenseVoice/Paraformer STT models

#790 Feature Request: Add support for Zyphra/ZONOS2 TTS model

#779 Support for Granite 4.1 - Standard and Plus variants?

#647 Feature request: STT support for FunAudioLLM/Fun-ASR-Nano-2512

#762 Whisper best_of-N sampling crashes: `tokens.reshape(tokens, ...)` passes the array as the shape arg (decoding.py:662-664)

#773 nemotron_asr: add cache-aware streaming (currently offline-only)