All pull requests

Pull requests

• speechlm2 vLLM: encoder CUDA graphs for audio of any length

  ASR
  #16389·pzelasko opened Oct 15, 2026Collaborator·4 of 8 tasks·

  ·
  1

• Add Riva Translate v2 support

  ASR
  #16388·briyan-nvidia opened Oct 15, 2026·1 of 8 tasks·

  ·

• feat(speechlm): support native RL checkpoints and frozen CTC export

  #16387·vmendelev opened Oct 15, 2026Collaborator·4 tasks done·

  ·
  4

• Serve waveform-to-text TDT recognition with TensorRT

  ASR
  #16386·albertz opened Oct 15, 2026Member·4 tasks done·

  ·
  6

• Fix corpus cpWER aggregation across sessions

  ASRcommunity-request
  #16385·tonydzi opened Oct 15, 2026·5 of 7 tasks·

  ·
  1

  5

• Merge #16279 and #16295 to main

  TTS
  #16384·blisc opened Oct 14, 2026Collaborator·

  ·
  2

• Merge #16301 to main

  TTS
  #16383·blisc opened Oct 14, 2026Collaborator·

  ·
  2

• Merge #16230 to main

  TTS
  #16382·blisc opened Oct 14, 2026Collaborator·

  ·
  2

• Merges #16294 into main

  TTS
  #16381·blisc opened Oct 14, 2026Collaborator·

  ·
  5

• fix(asr): count VAD false alarms outside the reference extent

  ASRcommunity-request
  #16380·Kayvan-Zahiri opened Oct 14, 2026Contributor·1 task done·

  ·
  1

• Merges #16238, #16242, and #16343 to main

  TTS
  #16379·blisc opened Oct 14, 2026Collaborator·

  ·
  1

• fix(asr): check the prediction's own type in frame_vad_construct_supervisions_per_file

  ASRcommunity-request
  #16378·Kayvan-Zahiri opened Oct 14, 2026Contributor·1 task done·

  ·
  1

• Preserve requested order in UTMOS scoring (#16266)

  TTS
  #16377·blisc opened Oct 14, 2026Collaborator·

  ·
  4

• Add complete TDT decoder component and Triton serving example

  ASR
  #16375·albertz opened Oct 14, 2026Member·4 tasks done·

  ·
  6

• Add direct Torch-TensorRT conversion for Transformer ASR encoders

  ASR
  #16374·albertz opened Oct 14, 2026Member·4 tasks done·

  ·
  5

• Add standalone TensorRT waveform preprocessing

  ASR
  #16373·albertz opened Oct 14, 2026Member·3 tasks done·

  ·
  6

• Add TensorRT builder for RNNT/TDT prediction and joint networks

  ASR
  #16372·albertz opened Oct 14, 2026Member·4 tasks done·

  ·
  3

• Add TensorRT execution for offline ASR encoders

  ASR
  #16371·albertz opened Oct 14, 2026Member·6 tasks done·

  ·
  8

• speechlm2 vLLM: CUDA graphs for the ParallelExpertEncoder

  ASRcommon
  #16369·pzelasko opened Oct 13, 2026Collaborator·4 of 8 tasks·

  ·
  24

• Enable offline Transformer encoder ONNX export

  ASR
  #16368·albertz opened Oct 13, 2026Member·4 tasks done·

  ·
  8

• [ASR] Run the chunked inference scripts on CPU when cuda is negative

  ASRcommunity-requestwaiting-on-maintainers
  #16367·ANDROIDHASSAN opened Oct 13, 2026·2 of 4 tasks·

  ·
  1

• [ASR] Run transcribe_speech.py on CPU when cuda is negative

  ASRcommunity-requestwaiting-on-maintainers
  #16365·ANDROIDHASSAN opened Oct 13, 2026·2 of 4 tasks·

  ·
  1

• Add GT-prefix multi-turn GRPO

  TTS
  #16364·shehzeen opened Oct 12, 2026Collaborator·

  ·
  1

• fix(ci): use main environment for scheduled failure alerts

  CI
  #16360·ko3n1g opened Oct 10, 2026Contributor·5 tasks done·

  ·
  1

• Fix agent skills: normalize unofficial frontmatter fields per the Agent Skills spec

  #16358·janEbert opened Oct 10, 2026Contributor·

  ·
  1